16cbd5570SChris Mason /* 26cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 36cbd5570SChris Mason * 46cbd5570SChris Mason * This program is free software; you can redistribute it and/or 56cbd5570SChris Mason * modify it under the terms of the GNU General Public 66cbd5570SChris Mason * License v2 as published by the Free Software Foundation. 76cbd5570SChris Mason * 86cbd5570SChris Mason * This program is distributed in the hope that it will be useful, 96cbd5570SChris Mason * but WITHOUT ANY WARRANTY; without even the implied warranty of 106cbd5570SChris Mason * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 116cbd5570SChris Mason * General Public License for more details. 126cbd5570SChris Mason * 136cbd5570SChris Mason * You should have received a copy of the GNU General Public 146cbd5570SChris Mason * License along with this program; if not, write to the 156cbd5570SChris Mason * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 166cbd5570SChris Mason * Boston, MA 021110-1307, USA. 176cbd5570SChris Mason */ 186cbd5570SChris Mason 19e20d96d6SChris Mason #include <linux/fs.h> 20d98237b3SChris Mason #include <linux/blkdev.h> 2187cbda5cSChris Mason #include <linux/scatterlist.h> 2222b0ebdaSChris Mason #include <linux/swap.h> 230f7d52f4SChris Mason #include <linux/radix-tree.h> 2435b7e476SChris Mason #include <linux/writeback.h> 25d397712bSChris Mason #include <linux/buffer_head.h> 26ce9adaa5SChris Mason #include <linux/workqueue.h> 27a74a4b97SChris Mason #include <linux/kthread.h> 28a74a4b97SChris Mason #include <linux/freezer.h> 29163e783eSDavid Woodhouse #include <linux/crc32c.h> 305a0e3ad6STejun Heo #include <linux/slab.h> 31784b4e29SChris Mason #include <linux/migrate.h> 327a36ddecSDavid Sterba #include <linux/ratelimit.h> 336463fe58SStefan Behrens #include <linux/uuid.h> 34803b2f54SStefan Behrens #include <linux/semaphore.h> 357e75bf3fSDavid Sterba #include <asm/unaligned.h> 364b4e25f2SChris Mason #include "compat.h" 37eb60ceacSChris Mason #include "ctree.h" 38eb60ceacSChris Mason #include "disk-io.h" 39e089f05cSChris Mason #include "transaction.h" 400f7d52f4SChris Mason #include "btrfs_inode.h" 410b86a832SChris Mason #include "volumes.h" 42db94535dSChris Mason #include "print-tree.h" 438b712842SChris Mason #include "async-thread.h" 44925baeddSChris Mason #include "locking.h" 45e02119d5SChris Mason #include "tree-log.h" 46fa9c0d79SChris Mason #include "free-space-cache.h" 47581bb050SLi Zefan #include "inode-map.h" 4821adbd5cSStefan Behrens #include "check-integrity.h" 49606686eeSJosef Bacik #include "rcu-string.h" 508dabb742SStefan Behrens #include "dev-replace.h" 5153b381b3SDavid Woodhouse #include "raid56.h" 52eb60ceacSChris Mason 53de0022b9SJosef Bacik #ifdef CONFIG_X86 54de0022b9SJosef Bacik #include <asm/cpufeature.h> 55de0022b9SJosef Bacik #endif 56de0022b9SJosef Bacik 57d1310b2eSChris Mason static struct extent_io_ops btree_extent_io_ops; 588b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work); 594df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root); 60fcd1f065SDavid Sterba static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 61acce952bSliubo int read_only); 62569e0f35SJosef Bacik static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t, 63569e0f35SJosef Bacik struct btrfs_root *root); 64143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 65acce952bSliubo static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 66acce952bSliubo struct btrfs_root *root); 67143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 68acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 69acce952bSliubo struct extent_io_tree *dirty_pages, 70acce952bSliubo int mark); 71acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 72acce952bSliubo struct extent_io_tree *pinned_extents); 7348a3b636SEric Sandeen static int btrfs_cleanup_transaction(struct btrfs_root *root); 7448a3b636SEric Sandeen static void btrfs_error_commit_super(struct btrfs_root *root); 75ce9adaa5SChris Mason 76d352ac68SChris Mason /* 77d352ac68SChris Mason * end_io_wq structs are used to do processing in task context when an IO is 78d352ac68SChris Mason * complete. This is used during reads to verify checksums, and it is used 79d352ac68SChris Mason * by writes to insert metadata for new file extents after IO is complete. 80d352ac68SChris Mason */ 81ce9adaa5SChris Mason struct end_io_wq { 82ce9adaa5SChris Mason struct bio *bio; 83ce9adaa5SChris Mason bio_end_io_t *end_io; 84ce9adaa5SChris Mason void *private; 85ce9adaa5SChris Mason struct btrfs_fs_info *info; 86ce9adaa5SChris Mason int error; 8722c59948SChris Mason int metadata; 88ce9adaa5SChris Mason struct list_head list; 898b712842SChris Mason struct btrfs_work work; 90ce9adaa5SChris Mason }; 910da5468fSChris Mason 92d352ac68SChris Mason /* 93d352ac68SChris Mason * async submit bios are used to offload expensive checksumming 94d352ac68SChris Mason * onto the worker threads. They checksum file and metadata bios 95d352ac68SChris Mason * just before they are sent down the IO stack. 96d352ac68SChris Mason */ 9744b8bd7eSChris Mason struct async_submit_bio { 9844b8bd7eSChris Mason struct inode *inode; 9944b8bd7eSChris Mason struct bio *bio; 10044b8bd7eSChris Mason struct list_head list; 1014a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start; 1024a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done; 10344b8bd7eSChris Mason int rw; 10444b8bd7eSChris Mason int mirror_num; 105c8b97818SChris Mason unsigned long bio_flags; 106eaf25d93SChris Mason /* 107eaf25d93SChris Mason * bio_offset is optional, can be used if the pages in the bio 108eaf25d93SChris Mason * can't tell us where in the file the bio should go 109eaf25d93SChris Mason */ 110eaf25d93SChris Mason u64 bio_offset; 1118b712842SChris Mason struct btrfs_work work; 11279787eaaSJeff Mahoney int error; 11344b8bd7eSChris Mason }; 11444b8bd7eSChris Mason 11585d4e461SChris Mason /* 11685d4e461SChris Mason * Lockdep class keys for extent_buffer->lock's in this root. For a given 11785d4e461SChris Mason * eb, the lockdep key is determined by the btrfs_root it belongs to and 11885d4e461SChris Mason * the level the eb occupies in the tree. 1194008c04aSChris Mason * 12085d4e461SChris Mason * Different roots are used for different purposes and may nest inside each 12185d4e461SChris Mason * other and they require separate keysets. As lockdep keys should be 12285d4e461SChris Mason * static, assign keysets according to the purpose of the root as indicated 12385d4e461SChris Mason * by btrfs_root->objectid. This ensures that all special purpose roots 12485d4e461SChris Mason * have separate keysets. 1254008c04aSChris Mason * 12685d4e461SChris Mason * Lock-nesting across peer nodes is always done with the immediate parent 12785d4e461SChris Mason * node locked thus preventing deadlock. As lockdep doesn't know this, use 12885d4e461SChris Mason * subclass to avoid triggering lockdep warning in such cases. 1294008c04aSChris Mason * 13085d4e461SChris Mason * The key is set by the readpage_end_io_hook after the buffer has passed 13185d4e461SChris Mason * csum validation but before the pages are unlocked. It is also set by 13285d4e461SChris Mason * btrfs_init_new_buffer on freshly allocated blocks. 13385d4e461SChris Mason * 13485d4e461SChris Mason * We also add a check to make sure the highest level of the tree is the 13585d4e461SChris Mason * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 13685d4e461SChris Mason * needs update as well. 1374008c04aSChris Mason */ 1384008c04aSChris Mason #ifdef CONFIG_DEBUG_LOCK_ALLOC 1394008c04aSChris Mason # if BTRFS_MAX_LEVEL != 8 1404008c04aSChris Mason # error 1414008c04aSChris Mason # endif 14285d4e461SChris Mason 14385d4e461SChris Mason static struct btrfs_lockdep_keyset { 14485d4e461SChris Mason u64 id; /* root objectid */ 14585d4e461SChris Mason const char *name_stem; /* lock name stem */ 14685d4e461SChris Mason char names[BTRFS_MAX_LEVEL + 1][20]; 14785d4e461SChris Mason struct lock_class_key keys[BTRFS_MAX_LEVEL + 1]; 14885d4e461SChris Mason } btrfs_lockdep_keysets[] = { 14985d4e461SChris Mason { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" }, 15085d4e461SChris Mason { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" }, 15185d4e461SChris Mason { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" }, 15285d4e461SChris Mason { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" }, 15385d4e461SChris Mason { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" }, 15485d4e461SChris Mason { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" }, 15560b62978SDavid Sterba { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" }, 15685d4e461SChris Mason { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" }, 15785d4e461SChris Mason { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" }, 15885d4e461SChris Mason { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" }, 15913fd8da9SDavid Sterba { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" }, 16085d4e461SChris Mason { .id = 0, .name_stem = "tree" }, 1614008c04aSChris Mason }; 16285d4e461SChris Mason 16385d4e461SChris Mason void __init btrfs_init_lockdep(void) 16485d4e461SChris Mason { 16585d4e461SChris Mason int i, j; 16685d4e461SChris Mason 16785d4e461SChris Mason /* initialize lockdep class names */ 16885d4e461SChris Mason for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) { 16985d4e461SChris Mason struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i]; 17085d4e461SChris Mason 17185d4e461SChris Mason for (j = 0; j < ARRAY_SIZE(ks->names); j++) 17285d4e461SChris Mason snprintf(ks->names[j], sizeof(ks->names[j]), 17385d4e461SChris Mason "btrfs-%s-%02d", ks->name_stem, j); 17485d4e461SChris Mason } 17585d4e461SChris Mason } 17685d4e461SChris Mason 17785d4e461SChris Mason void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 17885d4e461SChris Mason int level) 17985d4e461SChris Mason { 18085d4e461SChris Mason struct btrfs_lockdep_keyset *ks; 18185d4e461SChris Mason 18285d4e461SChris Mason BUG_ON(level >= ARRAY_SIZE(ks->keys)); 18385d4e461SChris Mason 18485d4e461SChris Mason /* find the matching keyset, id 0 is the default entry */ 18585d4e461SChris Mason for (ks = btrfs_lockdep_keysets; ks->id; ks++) 18685d4e461SChris Mason if (ks->id == objectid) 18785d4e461SChris Mason break; 18885d4e461SChris Mason 18985d4e461SChris Mason lockdep_set_class_and_name(&eb->lock, 19085d4e461SChris Mason &ks->keys[level], ks->names[level]); 19185d4e461SChris Mason } 19285d4e461SChris Mason 1934008c04aSChris Mason #endif 1944008c04aSChris Mason 195d352ac68SChris Mason /* 196d352ac68SChris Mason * extents on the btree inode are pretty simple, there's one extent 197d352ac68SChris Mason * that covers the entire device 198d352ac68SChris Mason */ 199b2950863SChristoph Hellwig static struct extent_map *btree_get_extent(struct inode *inode, 200306e16ceSDavid Sterba struct page *page, size_t pg_offset, u64 start, u64 len, 2015f39d397SChris Mason int create) 2025f39d397SChris Mason { 2035f39d397SChris Mason struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 2045f39d397SChris Mason struct extent_map *em; 2055f39d397SChris Mason int ret; 2065f39d397SChris Mason 207890871beSChris Mason read_lock(&em_tree->lock); 208d1310b2eSChris Mason em = lookup_extent_mapping(em_tree, start, len); 209a061fc8dSChris Mason if (em) { 210a061fc8dSChris Mason em->bdev = 211a061fc8dSChris Mason BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 212890871beSChris Mason read_unlock(&em_tree->lock); 2135f39d397SChris Mason goto out; 214a061fc8dSChris Mason } 215890871beSChris Mason read_unlock(&em_tree->lock); 2167b13b7b1SChris Mason 217172ddd60SDavid Sterba em = alloc_extent_map(); 2185f39d397SChris Mason if (!em) { 2195f39d397SChris Mason em = ERR_PTR(-ENOMEM); 2205f39d397SChris Mason goto out; 2215f39d397SChris Mason } 2225f39d397SChris Mason em->start = 0; 2230afbaf8cSChris Mason em->len = (u64)-1; 224c8b97818SChris Mason em->block_len = (u64)-1; 2255f39d397SChris Mason em->block_start = 0; 226a061fc8dSChris Mason em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 227d1310b2eSChris Mason 228890871beSChris Mason write_lock(&em_tree->lock); 22909a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 2305f39d397SChris Mason if (ret == -EEXIST) { 2315f39d397SChris Mason free_extent_map(em); 2327b13b7b1SChris Mason em = lookup_extent_mapping(em_tree, start, len); 233b4f359abSTsutomu Itoh if (!em) 2340433f20dSTsutomu Itoh em = ERR_PTR(-EIO); 2355f39d397SChris Mason } else if (ret) { 2367b13b7b1SChris Mason free_extent_map(em); 2370433f20dSTsutomu Itoh em = ERR_PTR(ret); 2385f39d397SChris Mason } 239890871beSChris Mason write_unlock(&em_tree->lock); 2407b13b7b1SChris Mason 2415f39d397SChris Mason out: 2425f39d397SChris Mason return em; 2435f39d397SChris Mason } 2445f39d397SChris Mason 245b0496686SLiu Bo u32 btrfs_csum_data(char *data, u32 seed, size_t len) 24619c00ddcSChris Mason { 247163e783eSDavid Woodhouse return crc32c(seed, data, len); 24819c00ddcSChris Mason } 24919c00ddcSChris Mason 25019c00ddcSChris Mason void btrfs_csum_final(u32 crc, char *result) 25119c00ddcSChris Mason { 2527e75bf3fSDavid Sterba put_unaligned_le32(~crc, result); 25319c00ddcSChris Mason } 25419c00ddcSChris Mason 255d352ac68SChris Mason /* 256d352ac68SChris Mason * compute the csum for a btree block, and either verify it or write it 257d352ac68SChris Mason * into the csum field of the block. 258d352ac68SChris Mason */ 25919c00ddcSChris Mason static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf, 26019c00ddcSChris Mason int verify) 26119c00ddcSChris Mason { 2626c41761fSDavid Sterba u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy); 263607d432dSJosef Bacik char *result = NULL; 26419c00ddcSChris Mason unsigned long len; 26519c00ddcSChris Mason unsigned long cur_len; 26619c00ddcSChris Mason unsigned long offset = BTRFS_CSUM_SIZE; 26719c00ddcSChris Mason char *kaddr; 26819c00ddcSChris Mason unsigned long map_start; 26919c00ddcSChris Mason unsigned long map_len; 27019c00ddcSChris Mason int err; 27119c00ddcSChris Mason u32 crc = ~(u32)0; 272607d432dSJosef Bacik unsigned long inline_result; 27319c00ddcSChris Mason 27419c00ddcSChris Mason len = buf->len - offset; 27519c00ddcSChris Mason while (len > 0) { 27619c00ddcSChris Mason err = map_private_extent_buffer(buf, offset, 32, 277a6591715SChris Mason &kaddr, &map_start, &map_len); 278d397712bSChris Mason if (err) 27919c00ddcSChris Mason return 1; 28019c00ddcSChris Mason cur_len = min(len, map_len - (offset - map_start)); 281b0496686SLiu Bo crc = btrfs_csum_data(kaddr + offset - map_start, 28219c00ddcSChris Mason crc, cur_len); 28319c00ddcSChris Mason len -= cur_len; 28419c00ddcSChris Mason offset += cur_len; 28519c00ddcSChris Mason } 286607d432dSJosef Bacik if (csum_size > sizeof(inline_result)) { 287607d432dSJosef Bacik result = kzalloc(csum_size * sizeof(char), GFP_NOFS); 288607d432dSJosef Bacik if (!result) 289607d432dSJosef Bacik return 1; 290607d432dSJosef Bacik } else { 291607d432dSJosef Bacik result = (char *)&inline_result; 292607d432dSJosef Bacik } 293607d432dSJosef Bacik 29419c00ddcSChris Mason btrfs_csum_final(crc, result); 29519c00ddcSChris Mason 29619c00ddcSChris Mason if (verify) { 297607d432dSJosef Bacik if (memcmp_extent_buffer(buf, result, 0, csum_size)) { 298e4204dedSChris Mason u32 val; 299e4204dedSChris Mason u32 found = 0; 300607d432dSJosef Bacik memcpy(&found, result, csum_size); 301e4204dedSChris Mason 302607d432dSJosef Bacik read_extent_buffer(buf, &val, 0, csum_size); 3037a36ddecSDavid Sterba printk_ratelimited(KERN_INFO "btrfs: %s checksum verify " 304193f284dSChris Mason "failed on %llu wanted %X found %X " 305193f284dSChris Mason "level %d\n", 306c1c9ff7cSGeert Uytterhoeven root->fs_info->sb->s_id, buf->start, 307c1c9ff7cSGeert Uytterhoeven val, found, btrfs_header_level(buf)); 308607d432dSJosef Bacik if (result != (char *)&inline_result) 309607d432dSJosef Bacik kfree(result); 31019c00ddcSChris Mason return 1; 31119c00ddcSChris Mason } 31219c00ddcSChris Mason } else { 313607d432dSJosef Bacik write_extent_buffer(buf, result, 0, csum_size); 31419c00ddcSChris Mason } 315607d432dSJosef Bacik if (result != (char *)&inline_result) 316607d432dSJosef Bacik kfree(result); 31719c00ddcSChris Mason return 0; 31819c00ddcSChris Mason } 31919c00ddcSChris Mason 320d352ac68SChris Mason /* 321d352ac68SChris Mason * we can't consider a given block up to date unless the transid of the 322d352ac68SChris Mason * block matches the transid in the parent node's pointer. This is how we 323d352ac68SChris Mason * detect blocks that either didn't get written at all or got written 324d352ac68SChris Mason * in the wrong place. 325d352ac68SChris Mason */ 3261259ab75SChris Mason static int verify_parent_transid(struct extent_io_tree *io_tree, 327b9fab919SChris Mason struct extent_buffer *eb, u64 parent_transid, 328b9fab919SChris Mason int atomic) 3291259ab75SChris Mason { 3302ac55d41SJosef Bacik struct extent_state *cached_state = NULL; 3311259ab75SChris Mason int ret; 3321259ab75SChris Mason 3331259ab75SChris Mason if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 3341259ab75SChris Mason return 0; 3351259ab75SChris Mason 336b9fab919SChris Mason if (atomic) 337b9fab919SChris Mason return -EAGAIN; 338b9fab919SChris Mason 3392ac55d41SJosef Bacik lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 340d0082371SJeff Mahoney 0, &cached_state); 3410b32f4bbSJosef Bacik if (extent_buffer_uptodate(eb) && 3421259ab75SChris Mason btrfs_header_generation(eb) == parent_transid) { 3431259ab75SChris Mason ret = 0; 3441259ab75SChris Mason goto out; 3451259ab75SChris Mason } 3467a36ddecSDavid Sterba printk_ratelimited("parent transid verify failed on %llu wanted %llu " 347193f284dSChris Mason "found %llu\n", 348c1c9ff7cSGeert Uytterhoeven eb->start, parent_transid, btrfs_header_generation(eb)); 3491259ab75SChris Mason ret = 1; 3500b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 35133958dc6SChris Mason out: 3522ac55d41SJosef Bacik unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 3532ac55d41SJosef Bacik &cached_state, GFP_NOFS); 3541259ab75SChris Mason return ret; 3551259ab75SChris Mason } 3561259ab75SChris Mason 357d352ac68SChris Mason /* 3581104a885SDavid Sterba * Return 0 if the superblock checksum type matches the checksum value of that 3591104a885SDavid Sterba * algorithm. Pass the raw disk superblock data. 3601104a885SDavid Sterba */ 3611104a885SDavid Sterba static int btrfs_check_super_csum(char *raw_disk_sb) 3621104a885SDavid Sterba { 3631104a885SDavid Sterba struct btrfs_super_block *disk_sb = 3641104a885SDavid Sterba (struct btrfs_super_block *)raw_disk_sb; 3651104a885SDavid Sterba u16 csum_type = btrfs_super_csum_type(disk_sb); 3661104a885SDavid Sterba int ret = 0; 3671104a885SDavid Sterba 3681104a885SDavid Sterba if (csum_type == BTRFS_CSUM_TYPE_CRC32) { 3691104a885SDavid Sterba u32 crc = ~(u32)0; 3701104a885SDavid Sterba const int csum_size = sizeof(crc); 3711104a885SDavid Sterba char result[csum_size]; 3721104a885SDavid Sterba 3731104a885SDavid Sterba /* 3741104a885SDavid Sterba * The super_block structure does not span the whole 3751104a885SDavid Sterba * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space 3761104a885SDavid Sterba * is filled with zeros and is included in the checkum. 3771104a885SDavid Sterba */ 3781104a885SDavid Sterba crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE, 3791104a885SDavid Sterba crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 3801104a885SDavid Sterba btrfs_csum_final(crc, result); 3811104a885SDavid Sterba 3821104a885SDavid Sterba if (memcmp(raw_disk_sb, result, csum_size)) 3831104a885SDavid Sterba ret = 1; 384667e7d94SChris Mason 385667e7d94SChris Mason if (ret && btrfs_super_generation(disk_sb) < 10) { 386667e7d94SChris Mason printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n"); 387667e7d94SChris Mason ret = 0; 388667e7d94SChris Mason } 3891104a885SDavid Sterba } 3901104a885SDavid Sterba 3911104a885SDavid Sterba if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) { 3921104a885SDavid Sterba printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n", 3931104a885SDavid Sterba csum_type); 3941104a885SDavid Sterba ret = 1; 3951104a885SDavid Sterba } 3961104a885SDavid Sterba 3971104a885SDavid Sterba return ret; 3981104a885SDavid Sterba } 3991104a885SDavid Sterba 4001104a885SDavid Sterba /* 401d352ac68SChris Mason * helper to read a given tree block, doing retries as required when 402d352ac68SChris Mason * the checksums don't match and we have alternate mirrors to try. 403d352ac68SChris Mason */ 404f188591eSChris Mason static int btree_read_extent_buffer_pages(struct btrfs_root *root, 405f188591eSChris Mason struct extent_buffer *eb, 406ca7a79adSChris Mason u64 start, u64 parent_transid) 407f188591eSChris Mason { 408f188591eSChris Mason struct extent_io_tree *io_tree; 409ea466794SJosef Bacik int failed = 0; 410f188591eSChris Mason int ret; 411f188591eSChris Mason int num_copies = 0; 412f188591eSChris Mason int mirror_num = 0; 413ea466794SJosef Bacik int failed_mirror = 0; 414f188591eSChris Mason 415a826d6dcSJosef Bacik clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 416f188591eSChris Mason io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree; 417f188591eSChris Mason while (1) { 418bb82ab88SArne Jansen ret = read_extent_buffer_pages(io_tree, eb, start, 419bb82ab88SArne Jansen WAIT_COMPLETE, 420f188591eSChris Mason btree_get_extent, mirror_num); 421256dd1bbSStefan Behrens if (!ret) { 422256dd1bbSStefan Behrens if (!verify_parent_transid(io_tree, eb, 423b9fab919SChris Mason parent_transid, 0)) 424ea466794SJosef Bacik break; 425256dd1bbSStefan Behrens else 426256dd1bbSStefan Behrens ret = -EIO; 427256dd1bbSStefan Behrens } 428d397712bSChris Mason 429a826d6dcSJosef Bacik /* 430a826d6dcSJosef Bacik * This buffer's crc is fine, but its contents are corrupted, so 431a826d6dcSJosef Bacik * there is no reason to read the other copies, they won't be 432a826d6dcSJosef Bacik * any less wrong. 433a826d6dcSJosef Bacik */ 434a826d6dcSJosef Bacik if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags)) 435ea466794SJosef Bacik break; 436ea466794SJosef Bacik 4375d964051SStefan Behrens num_copies = btrfs_num_copies(root->fs_info, 438f188591eSChris Mason eb->start, eb->len); 4394235298eSChris Mason if (num_copies == 1) 440ea466794SJosef Bacik break; 4414235298eSChris Mason 4425cf1ab56SJosef Bacik if (!failed_mirror) { 4435cf1ab56SJosef Bacik failed = 1; 4445cf1ab56SJosef Bacik failed_mirror = eb->read_mirror; 4455cf1ab56SJosef Bacik } 4465cf1ab56SJosef Bacik 447f188591eSChris Mason mirror_num++; 448ea466794SJosef Bacik if (mirror_num == failed_mirror) 449ea466794SJosef Bacik mirror_num++; 450ea466794SJosef Bacik 4514235298eSChris Mason if (mirror_num > num_copies) 452ea466794SJosef Bacik break; 453f188591eSChris Mason } 454ea466794SJosef Bacik 455c0901581SStefan Behrens if (failed && !ret && failed_mirror) 456ea466794SJosef Bacik repair_eb_io_failure(root, eb, failed_mirror); 457ea466794SJosef Bacik 458ea466794SJosef Bacik return ret; 459f188591eSChris Mason } 46019c00ddcSChris Mason 461d352ac68SChris Mason /* 462d397712bSChris Mason * checksum a dirty tree block before IO. This has extra checks to make sure 463d397712bSChris Mason * we only fill in the checksum field in the first page of a multi-page block 464d352ac68SChris Mason */ 465d397712bSChris Mason 466b2950863SChristoph Hellwig static int csum_dirty_buffer(struct btrfs_root *root, struct page *page) 46719c00ddcSChris Mason { 468d1310b2eSChris Mason struct extent_io_tree *tree; 4694eee4fa4SMiao Xie u64 start = page_offset(page); 47019c00ddcSChris Mason u64 found_start; 47119c00ddcSChris Mason struct extent_buffer *eb; 472f188591eSChris Mason 473d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 47419c00ddcSChris Mason 4754f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 4764f2de97aSJosef Bacik if (page != eb->pages[0]) 4774f2de97aSJosef Bacik return 0; 47819c00ddcSChris Mason found_start = btrfs_header_bytenr(eb); 47919c00ddcSChris Mason if (found_start != start) { 48055c69072SChris Mason WARN_ON(1); 4814f2de97aSJosef Bacik return 0; 48255c69072SChris Mason } 48355c69072SChris Mason if (!PageUptodate(page)) { 48455c69072SChris Mason WARN_ON(1); 4854f2de97aSJosef Bacik return 0; 48619c00ddcSChris Mason } 48719c00ddcSChris Mason csum_tree_block(root, eb, 0); 48819c00ddcSChris Mason return 0; 48919c00ddcSChris Mason } 49019c00ddcSChris Mason 4912b82032cSYan Zheng static int check_tree_block_fsid(struct btrfs_root *root, 4922b82032cSYan Zheng struct extent_buffer *eb) 4932b82032cSYan Zheng { 4942b82032cSYan Zheng struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 4952b82032cSYan Zheng u8 fsid[BTRFS_UUID_SIZE]; 4962b82032cSYan Zheng int ret = 1; 4972b82032cSYan Zheng 4980a4e5586SRoss Kirk read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 4992b82032cSYan Zheng while (fs_devices) { 5002b82032cSYan Zheng if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) { 5012b82032cSYan Zheng ret = 0; 5022b82032cSYan Zheng break; 5032b82032cSYan Zheng } 5042b82032cSYan Zheng fs_devices = fs_devices->seed; 5052b82032cSYan Zheng } 5062b82032cSYan Zheng return ret; 5072b82032cSYan Zheng } 5082b82032cSYan Zheng 509a826d6dcSJosef Bacik #define CORRUPT(reason, eb, root, slot) \ 510a826d6dcSJosef Bacik printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \ 511a826d6dcSJosef Bacik "root=%llu, slot=%d\n", reason, \ 512c1c9ff7cSGeert Uytterhoeven btrfs_header_bytenr(eb), root->objectid, slot) 513a826d6dcSJosef Bacik 514a826d6dcSJosef Bacik static noinline int check_leaf(struct btrfs_root *root, 515a826d6dcSJosef Bacik struct extent_buffer *leaf) 516a826d6dcSJosef Bacik { 517a826d6dcSJosef Bacik struct btrfs_key key; 518a826d6dcSJosef Bacik struct btrfs_key leaf_key; 519a826d6dcSJosef Bacik u32 nritems = btrfs_header_nritems(leaf); 520a826d6dcSJosef Bacik int slot; 521a826d6dcSJosef Bacik 522a826d6dcSJosef Bacik if (nritems == 0) 523a826d6dcSJosef Bacik return 0; 524a826d6dcSJosef Bacik 525a826d6dcSJosef Bacik /* Check the 0 item */ 526a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) != 527a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 528a826d6dcSJosef Bacik CORRUPT("invalid item offset size pair", leaf, root, 0); 529a826d6dcSJosef Bacik return -EIO; 530a826d6dcSJosef Bacik } 531a826d6dcSJosef Bacik 532a826d6dcSJosef Bacik /* 533a826d6dcSJosef Bacik * Check to make sure each items keys are in the correct order and their 534a826d6dcSJosef Bacik * offsets make sense. We only have to loop through nritems-1 because 535a826d6dcSJosef Bacik * we check the current slot against the next slot, which verifies the 536a826d6dcSJosef Bacik * next slot's offset+size makes sense and that the current's slot 537a826d6dcSJosef Bacik * offset is correct. 538a826d6dcSJosef Bacik */ 539a826d6dcSJosef Bacik for (slot = 0; slot < nritems - 1; slot++) { 540a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &leaf_key, slot); 541a826d6dcSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot + 1); 542a826d6dcSJosef Bacik 543a826d6dcSJosef Bacik /* Make sure the keys are in the right order */ 544a826d6dcSJosef Bacik if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) { 545a826d6dcSJosef Bacik CORRUPT("bad key order", leaf, root, slot); 546a826d6dcSJosef Bacik return -EIO; 547a826d6dcSJosef Bacik } 548a826d6dcSJosef Bacik 549a826d6dcSJosef Bacik /* 550a826d6dcSJosef Bacik * Make sure the offset and ends are right, remember that the 551a826d6dcSJosef Bacik * item data starts at the end of the leaf and grows towards the 552a826d6dcSJosef Bacik * front. 553a826d6dcSJosef Bacik */ 554a826d6dcSJosef Bacik if (btrfs_item_offset_nr(leaf, slot) != 555a826d6dcSJosef Bacik btrfs_item_end_nr(leaf, slot + 1)) { 556a826d6dcSJosef Bacik CORRUPT("slot offset bad", leaf, root, slot); 557a826d6dcSJosef Bacik return -EIO; 558a826d6dcSJosef Bacik } 559a826d6dcSJosef Bacik 560a826d6dcSJosef Bacik /* 561a826d6dcSJosef Bacik * Check to make sure that we don't point outside of the leaf, 562a826d6dcSJosef Bacik * just incase all the items are consistent to eachother, but 563a826d6dcSJosef Bacik * all point outside of the leaf. 564a826d6dcSJosef Bacik */ 565a826d6dcSJosef Bacik if (btrfs_item_end_nr(leaf, slot) > 566a826d6dcSJosef Bacik BTRFS_LEAF_DATA_SIZE(root)) { 567a826d6dcSJosef Bacik CORRUPT("slot end outside of leaf", leaf, root, slot); 568a826d6dcSJosef Bacik return -EIO; 569a826d6dcSJosef Bacik } 570a826d6dcSJosef Bacik } 571a826d6dcSJosef Bacik 572a826d6dcSJosef Bacik return 0; 573a826d6dcSJosef Bacik } 574a826d6dcSJosef Bacik 575facc8a22SMiao Xie static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 576facc8a22SMiao Xie u64 phy_offset, struct page *page, 577facc8a22SMiao Xie u64 start, u64 end, int mirror) 578ce9adaa5SChris Mason { 579ce9adaa5SChris Mason struct extent_io_tree *tree; 580ce9adaa5SChris Mason u64 found_start; 581ce9adaa5SChris Mason int found_level; 582ce9adaa5SChris Mason struct extent_buffer *eb; 583ce9adaa5SChris Mason struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 584f188591eSChris Mason int ret = 0; 585727011e0SChris Mason int reads_done; 586ce9adaa5SChris Mason 587ce9adaa5SChris Mason if (!page->private) 588ce9adaa5SChris Mason goto out; 589d397712bSChris Mason 590727011e0SChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 5914f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 592d397712bSChris Mason 5930b32f4bbSJosef Bacik /* the pending IO might have been the only thing that kept this buffer 5940b32f4bbSJosef Bacik * in memory. Make sure we have a ref for all this other checks 5950b32f4bbSJosef Bacik */ 5960b32f4bbSJosef Bacik extent_buffer_get(eb); 5970b32f4bbSJosef Bacik 5980b32f4bbSJosef Bacik reads_done = atomic_dec_and_test(&eb->io_pages); 599727011e0SChris Mason if (!reads_done) 600727011e0SChris Mason goto err; 601f188591eSChris Mason 6025cf1ab56SJosef Bacik eb->read_mirror = mirror; 603ea466794SJosef Bacik if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) { 604ea466794SJosef Bacik ret = -EIO; 605ea466794SJosef Bacik goto err; 606ea466794SJosef Bacik } 607ea466794SJosef Bacik 608ce9adaa5SChris Mason found_start = btrfs_header_bytenr(eb); 609727011e0SChris Mason if (found_start != eb->start) { 6107a36ddecSDavid Sterba printk_ratelimited(KERN_INFO "btrfs bad tree block start " 611193f284dSChris Mason "%llu %llu\n", 612c1c9ff7cSGeert Uytterhoeven found_start, eb->start); 613f188591eSChris Mason ret = -EIO; 614ce9adaa5SChris Mason goto err; 615ce9adaa5SChris Mason } 6162b82032cSYan Zheng if (check_tree_block_fsid(root, eb)) { 6177a36ddecSDavid Sterba printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n", 618c1c9ff7cSGeert Uytterhoeven eb->start); 6191259ab75SChris Mason ret = -EIO; 6201259ab75SChris Mason goto err; 6211259ab75SChris Mason } 622ce9adaa5SChris Mason found_level = btrfs_header_level(eb); 6231c24c3ceSJosef Bacik if (found_level >= BTRFS_MAX_LEVEL) { 6241c24c3ceSJosef Bacik btrfs_info(root->fs_info, "bad tree block level %d\n", 6251c24c3ceSJosef Bacik (int)btrfs_header_level(eb)); 6261c24c3ceSJosef Bacik ret = -EIO; 6271c24c3ceSJosef Bacik goto err; 6281c24c3ceSJosef Bacik } 629ce9adaa5SChris Mason 63085d4e461SChris Mason btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 63185d4e461SChris Mason eb, found_level); 6324008c04aSChris Mason 633ce9adaa5SChris Mason ret = csum_tree_block(root, eb, 1); 634a826d6dcSJosef Bacik if (ret) { 635f188591eSChris Mason ret = -EIO; 636a826d6dcSJosef Bacik goto err; 637a826d6dcSJosef Bacik } 638a826d6dcSJosef Bacik 639a826d6dcSJosef Bacik /* 640a826d6dcSJosef Bacik * If this is a leaf block and it is corrupt, set the corrupt bit so 641a826d6dcSJosef Bacik * that we don't try and read the other copies of this block, just 642a826d6dcSJosef Bacik * return -EIO. 643a826d6dcSJosef Bacik */ 644a826d6dcSJosef Bacik if (found_level == 0 && check_leaf(root, eb)) { 645a826d6dcSJosef Bacik set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 646a826d6dcSJosef Bacik ret = -EIO; 647a826d6dcSJosef Bacik } 648ce9adaa5SChris Mason 6490b32f4bbSJosef Bacik if (!ret) 6500b32f4bbSJosef Bacik set_extent_buffer_uptodate(eb); 651ce9adaa5SChris Mason err: 65279fb65a1SJosef Bacik if (reads_done && 65379fb65a1SJosef Bacik test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 6544bb31e92SArne Jansen btree_readahead_hook(root, eb, eb->start, ret); 6554bb31e92SArne Jansen 65653b381b3SDavid Woodhouse if (ret) { 65753b381b3SDavid Woodhouse /* 65853b381b3SDavid Woodhouse * our io error hook is going to dec the io pages 65953b381b3SDavid Woodhouse * again, we have to make sure it has something 66053b381b3SDavid Woodhouse * to decrement 66153b381b3SDavid Woodhouse */ 66253b381b3SDavid Woodhouse atomic_inc(&eb->io_pages); 6630b32f4bbSJosef Bacik clear_extent_buffer_uptodate(eb); 66453b381b3SDavid Woodhouse } 6650b32f4bbSJosef Bacik free_extent_buffer(eb); 666ce9adaa5SChris Mason out: 667f188591eSChris Mason return ret; 668ce9adaa5SChris Mason } 669ce9adaa5SChris Mason 670ea466794SJosef Bacik static int btree_io_failed_hook(struct page *page, int failed_mirror) 6714bb31e92SArne Jansen { 6724bb31e92SArne Jansen struct extent_buffer *eb; 6734bb31e92SArne Jansen struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 6744bb31e92SArne Jansen 6754f2de97aSJosef Bacik eb = (struct extent_buffer *)page->private; 676ea466794SJosef Bacik set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); 6775cf1ab56SJosef Bacik eb->read_mirror = failed_mirror; 67853b381b3SDavid Woodhouse atomic_dec(&eb->io_pages); 679ea466794SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 6804bb31e92SArne Jansen btree_readahead_hook(root, eb, eb->start, -EIO); 6814bb31e92SArne Jansen return -EIO; /* we fixed nothing */ 6824bb31e92SArne Jansen } 6834bb31e92SArne Jansen 684ce9adaa5SChris Mason static void end_workqueue_bio(struct bio *bio, int err) 685ce9adaa5SChris Mason { 686ce9adaa5SChris Mason struct end_io_wq *end_io_wq = bio->bi_private; 687ce9adaa5SChris Mason struct btrfs_fs_info *fs_info; 688ce9adaa5SChris Mason 689ce9adaa5SChris Mason fs_info = end_io_wq->info; 690ce9adaa5SChris Mason end_io_wq->error = err; 6918b712842SChris Mason end_io_wq->work.func = end_workqueue_fn; 6928b712842SChris Mason end_io_wq->work.flags = 0; 693d20f7043SChris Mason 6947b6d91daSChristoph Hellwig if (bio->bi_rw & REQ_WRITE) { 69553b381b3SDavid Woodhouse if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 696cad321adSChris Mason btrfs_queue_worker(&fs_info->endio_meta_write_workers, 697cad321adSChris Mason &end_io_wq->work); 69853b381b3SDavid Woodhouse else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 6990cb59c99SJosef Bacik btrfs_queue_worker(&fs_info->endio_freespace_worker, 7000cb59c99SJosef Bacik &end_io_wq->work); 70153b381b3SDavid Woodhouse else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 70253b381b3SDavid Woodhouse btrfs_queue_worker(&fs_info->endio_raid56_workers, 70353b381b3SDavid Woodhouse &end_io_wq->work); 704cad321adSChris Mason else 705e6dcd2dcSChris Mason btrfs_queue_worker(&fs_info->endio_write_workers, 706e6dcd2dcSChris Mason &end_io_wq->work); 707d20f7043SChris Mason } else { 70853b381b3SDavid Woodhouse if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 70953b381b3SDavid Woodhouse btrfs_queue_worker(&fs_info->endio_raid56_workers, 71053b381b3SDavid Woodhouse &end_io_wq->work); 71153b381b3SDavid Woodhouse else if (end_io_wq->metadata) 712d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_meta_workers, 713d20f7043SChris Mason &end_io_wq->work); 714e6dcd2dcSChris Mason else 715d20f7043SChris Mason btrfs_queue_worker(&fs_info->endio_workers, 716d20f7043SChris Mason &end_io_wq->work); 717d20f7043SChris Mason } 718ce9adaa5SChris Mason } 719ce9adaa5SChris Mason 7200cb59c99SJosef Bacik /* 7210cb59c99SJosef Bacik * For the metadata arg you want 7220cb59c99SJosef Bacik * 7230cb59c99SJosef Bacik * 0 - if data 7240cb59c99SJosef Bacik * 1 - if normal metadta 7250cb59c99SJosef Bacik * 2 - if writing to the free space cache area 72653b381b3SDavid Woodhouse * 3 - raid parity work 7270cb59c99SJosef Bacik */ 72822c59948SChris Mason int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 72922c59948SChris Mason int metadata) 7300b86a832SChris Mason { 731ce9adaa5SChris Mason struct end_io_wq *end_io_wq; 732ce9adaa5SChris Mason end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS); 733ce9adaa5SChris Mason if (!end_io_wq) 734ce9adaa5SChris Mason return -ENOMEM; 735ce9adaa5SChris Mason 736ce9adaa5SChris Mason end_io_wq->private = bio->bi_private; 737ce9adaa5SChris Mason end_io_wq->end_io = bio->bi_end_io; 73822c59948SChris Mason end_io_wq->info = info; 739ce9adaa5SChris Mason end_io_wq->error = 0; 740ce9adaa5SChris Mason end_io_wq->bio = bio; 74122c59948SChris Mason end_io_wq->metadata = metadata; 742ce9adaa5SChris Mason 743ce9adaa5SChris Mason bio->bi_private = end_io_wq; 744ce9adaa5SChris Mason bio->bi_end_io = end_workqueue_bio; 74522c59948SChris Mason return 0; 74622c59948SChris Mason } 74722c59948SChris Mason 748b64a2851SChris Mason unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info) 7494854ddd0SChris Mason { 7504854ddd0SChris Mason unsigned long limit = min_t(unsigned long, 7514854ddd0SChris Mason info->workers.max_workers, 7524854ddd0SChris Mason info->fs_devices->open_devices); 7534854ddd0SChris Mason return 256 * limit; 7544854ddd0SChris Mason } 7554854ddd0SChris Mason 7564a69a410SChris Mason static void run_one_async_start(struct btrfs_work *work) 7574a69a410SChris Mason { 7584a69a410SChris Mason struct async_submit_bio *async; 75979787eaaSJeff Mahoney int ret; 7604a69a410SChris Mason 7614a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 76279787eaaSJeff Mahoney ret = async->submit_bio_start(async->inode, async->rw, async->bio, 763eaf25d93SChris Mason async->mirror_num, async->bio_flags, 764eaf25d93SChris Mason async->bio_offset); 76579787eaaSJeff Mahoney if (ret) 76679787eaaSJeff Mahoney async->error = ret; 7674a69a410SChris Mason } 7684a69a410SChris Mason 7694a69a410SChris Mason static void run_one_async_done(struct btrfs_work *work) 7708b712842SChris Mason { 7718b712842SChris Mason struct btrfs_fs_info *fs_info; 7728b712842SChris Mason struct async_submit_bio *async; 7734854ddd0SChris Mason int limit; 7748b712842SChris Mason 7758b712842SChris Mason async = container_of(work, struct async_submit_bio, work); 7768b712842SChris Mason fs_info = BTRFS_I(async->inode)->root->fs_info; 7774854ddd0SChris Mason 778b64a2851SChris Mason limit = btrfs_async_submit_limit(fs_info); 7794854ddd0SChris Mason limit = limit * 2 / 3; 7804854ddd0SChris Mason 78166657b31SJosef Bacik if (atomic_dec_return(&fs_info->nr_async_submits) < limit && 782b64a2851SChris Mason waitqueue_active(&fs_info->async_submit_wait)) 7834854ddd0SChris Mason wake_up(&fs_info->async_submit_wait); 7844854ddd0SChris Mason 78579787eaaSJeff Mahoney /* If an error occured we just want to clean up the bio and move on */ 78679787eaaSJeff Mahoney if (async->error) { 78779787eaaSJeff Mahoney bio_endio(async->bio, async->error); 78879787eaaSJeff Mahoney return; 78979787eaaSJeff Mahoney } 79079787eaaSJeff Mahoney 7914a69a410SChris Mason async->submit_bio_done(async->inode, async->rw, async->bio, 792eaf25d93SChris Mason async->mirror_num, async->bio_flags, 793eaf25d93SChris Mason async->bio_offset); 7944a69a410SChris Mason } 7954a69a410SChris Mason 7964a69a410SChris Mason static void run_one_async_free(struct btrfs_work *work) 7974a69a410SChris Mason { 7984a69a410SChris Mason struct async_submit_bio *async; 7994a69a410SChris Mason 8004a69a410SChris Mason async = container_of(work, struct async_submit_bio, work); 8018b712842SChris Mason kfree(async); 8028b712842SChris Mason } 8038b712842SChris Mason 80444b8bd7eSChris Mason int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode, 80544b8bd7eSChris Mason int rw, struct bio *bio, int mirror_num, 806c8b97818SChris Mason unsigned long bio_flags, 807eaf25d93SChris Mason u64 bio_offset, 8084a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_start, 8094a69a410SChris Mason extent_submit_bio_hook_t *submit_bio_done) 81044b8bd7eSChris Mason { 81144b8bd7eSChris Mason struct async_submit_bio *async; 81244b8bd7eSChris Mason 81344b8bd7eSChris Mason async = kmalloc(sizeof(*async), GFP_NOFS); 81444b8bd7eSChris Mason if (!async) 81544b8bd7eSChris Mason return -ENOMEM; 81644b8bd7eSChris Mason 81744b8bd7eSChris Mason async->inode = inode; 81844b8bd7eSChris Mason async->rw = rw; 81944b8bd7eSChris Mason async->bio = bio; 82044b8bd7eSChris Mason async->mirror_num = mirror_num; 8214a69a410SChris Mason async->submit_bio_start = submit_bio_start; 8224a69a410SChris Mason async->submit_bio_done = submit_bio_done; 8234a69a410SChris Mason 8244a69a410SChris Mason async->work.func = run_one_async_start; 8254a69a410SChris Mason async->work.ordered_func = run_one_async_done; 8264a69a410SChris Mason async->work.ordered_free = run_one_async_free; 8274a69a410SChris Mason 8288b712842SChris Mason async->work.flags = 0; 829c8b97818SChris Mason async->bio_flags = bio_flags; 830eaf25d93SChris Mason async->bio_offset = bio_offset; 8318c8bee1dSChris Mason 83279787eaaSJeff Mahoney async->error = 0; 83379787eaaSJeff Mahoney 834cb03c743SChris Mason atomic_inc(&fs_info->nr_async_submits); 835d313d7a3SChris Mason 8367b6d91daSChristoph Hellwig if (rw & REQ_SYNC) 837d313d7a3SChris Mason btrfs_set_work_high_prio(&async->work); 838d313d7a3SChris Mason 8398b712842SChris Mason btrfs_queue_worker(&fs_info->workers, &async->work); 8409473f16cSChris Mason 841771ed689SChris Mason while (atomic_read(&fs_info->async_submit_draining) && 842771ed689SChris Mason atomic_read(&fs_info->nr_async_submits)) { 843771ed689SChris Mason wait_event(fs_info->async_submit_wait, 844771ed689SChris Mason (atomic_read(&fs_info->nr_async_submits) == 0)); 845771ed689SChris Mason } 846771ed689SChris Mason 84744b8bd7eSChris Mason return 0; 84844b8bd7eSChris Mason } 84944b8bd7eSChris Mason 850ce3ed71aSChris Mason static int btree_csum_one_bio(struct bio *bio) 851ce3ed71aSChris Mason { 852ce3ed71aSChris Mason struct bio_vec *bvec = bio->bi_io_vec; 853ce3ed71aSChris Mason int bio_index = 0; 854ce3ed71aSChris Mason struct btrfs_root *root; 85579787eaaSJeff Mahoney int ret = 0; 856ce3ed71aSChris Mason 857ce3ed71aSChris Mason WARN_ON(bio->bi_vcnt <= 0); 858ce3ed71aSChris Mason while (bio_index < bio->bi_vcnt) { 859ce3ed71aSChris Mason root = BTRFS_I(bvec->bv_page->mapping->host)->root; 86079787eaaSJeff Mahoney ret = csum_dirty_buffer(root, bvec->bv_page); 86179787eaaSJeff Mahoney if (ret) 86279787eaaSJeff Mahoney break; 863ce3ed71aSChris Mason bio_index++; 864ce3ed71aSChris Mason bvec++; 865ce3ed71aSChris Mason } 86679787eaaSJeff Mahoney return ret; 867ce3ed71aSChris Mason } 868ce3ed71aSChris Mason 8694a69a410SChris Mason static int __btree_submit_bio_start(struct inode *inode, int rw, 8704a69a410SChris Mason struct bio *bio, int mirror_num, 871eaf25d93SChris Mason unsigned long bio_flags, 872eaf25d93SChris Mason u64 bio_offset) 87322c59948SChris Mason { 8748b712842SChris Mason /* 8758b712842SChris Mason * when we're called for a write, we're already in the async 8765443be45SChris Mason * submission context. Just jump into btrfs_map_bio 8778b712842SChris Mason */ 87879787eaaSJeff Mahoney return btree_csum_one_bio(bio); 87922c59948SChris Mason } 88022c59948SChris Mason 8814a69a410SChris Mason static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 882eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 883eaf25d93SChris Mason u64 bio_offset) 8844a69a410SChris Mason { 88561891923SStefan Behrens int ret; 88661891923SStefan Behrens 8878b712842SChris Mason /* 8884a69a410SChris Mason * when we're called for a write, we're already in the async 8894a69a410SChris Mason * submission context. Just jump into btrfs_map_bio 8908b712842SChris Mason */ 89161891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1); 89261891923SStefan Behrens if (ret) 89361891923SStefan Behrens bio_endio(bio, ret); 89461891923SStefan Behrens return ret; 8950b86a832SChris Mason } 8960b86a832SChris Mason 897de0022b9SJosef Bacik static int check_async_write(struct inode *inode, unsigned long bio_flags) 898de0022b9SJosef Bacik { 899de0022b9SJosef Bacik if (bio_flags & EXTENT_BIO_TREE_LOG) 900de0022b9SJosef Bacik return 0; 901de0022b9SJosef Bacik #ifdef CONFIG_X86 902de0022b9SJosef Bacik if (cpu_has_xmm4_2) 903de0022b9SJosef Bacik return 0; 904de0022b9SJosef Bacik #endif 905de0022b9SJosef Bacik return 1; 906de0022b9SJosef Bacik } 907de0022b9SJosef Bacik 90844b8bd7eSChris Mason static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 909eaf25d93SChris Mason int mirror_num, unsigned long bio_flags, 910eaf25d93SChris Mason u64 bio_offset) 91144b8bd7eSChris Mason { 912de0022b9SJosef Bacik int async = check_async_write(inode, bio_flags); 9134a69a410SChris Mason int ret; 914cad321adSChris Mason 9157b6d91daSChristoph Hellwig if (!(rw & REQ_WRITE)) { 916cad321adSChris Mason /* 917cad321adSChris Mason * called for a read, do the setup so that checksum validation 918cad321adSChris Mason * can happen in the async kernel threads 919cad321adSChris Mason */ 920f3f266abSChris Mason ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, 921f3f266abSChris Mason bio, 1); 9221d4284bdSChris Mason if (ret) 92361891923SStefan Behrens goto out_w_error; 92461891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 9256f3577bdSChris Mason mirror_num, 0); 926de0022b9SJosef Bacik } else if (!async) { 927de0022b9SJosef Bacik ret = btree_csum_one_bio(bio); 928de0022b9SJosef Bacik if (ret) 92961891923SStefan Behrens goto out_w_error; 93061891923SStefan Behrens ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 931de0022b9SJosef Bacik mirror_num, 0); 93261891923SStefan Behrens } else { 933cad321adSChris Mason /* 93461891923SStefan Behrens * kthread helpers are used to submit writes so that 93561891923SStefan Behrens * checksumming can happen in parallel across all CPUs 936cad321adSChris Mason */ 93761891923SStefan Behrens ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 938c8b97818SChris Mason inode, rw, bio, mirror_num, 0, 939eaf25d93SChris Mason bio_offset, 9404a69a410SChris Mason __btree_submit_bio_start, 9414a69a410SChris Mason __btree_submit_bio_done); 94244b8bd7eSChris Mason } 94344b8bd7eSChris Mason 94461891923SStefan Behrens if (ret) { 94561891923SStefan Behrens out_w_error: 94661891923SStefan Behrens bio_endio(bio, ret); 94761891923SStefan Behrens } 94861891923SStefan Behrens return ret; 94961891923SStefan Behrens } 95061891923SStefan Behrens 9513dd1462eSJan Beulich #ifdef CONFIG_MIGRATION 952784b4e29SChris Mason static int btree_migratepage(struct address_space *mapping, 953a6bc32b8SMel Gorman struct page *newpage, struct page *page, 954a6bc32b8SMel Gorman enum migrate_mode mode) 955784b4e29SChris Mason { 956784b4e29SChris Mason /* 957784b4e29SChris Mason * we can't safely write a btree page from here, 958784b4e29SChris Mason * we haven't done the locking hook 959784b4e29SChris Mason */ 960784b4e29SChris Mason if (PageDirty(page)) 961784b4e29SChris Mason return -EAGAIN; 962784b4e29SChris Mason /* 963784b4e29SChris Mason * Buffers may be managed in a filesystem specific way. 964784b4e29SChris Mason * We must have no buffers or drop them. 965784b4e29SChris Mason */ 966784b4e29SChris Mason if (page_has_private(page) && 967784b4e29SChris Mason !try_to_release_page(page, GFP_KERNEL)) 968784b4e29SChris Mason return -EAGAIN; 969a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 970784b4e29SChris Mason } 9713dd1462eSJan Beulich #endif 972784b4e29SChris Mason 9730da5468fSChris Mason 9740da5468fSChris Mason static int btree_writepages(struct address_space *mapping, 9750da5468fSChris Mason struct writeback_control *wbc) 9760da5468fSChris Mason { 977d1310b2eSChris Mason struct extent_io_tree *tree; 978e2d84521SMiao Xie struct btrfs_fs_info *fs_info; 979e2d84521SMiao Xie int ret; 980e2d84521SMiao Xie 981d1310b2eSChris Mason tree = &BTRFS_I(mapping->host)->io_tree; 982d8d5f3e1SChris Mason if (wbc->sync_mode == WB_SYNC_NONE) { 983448d640bSChris Mason 984448d640bSChris Mason if (wbc->for_kupdate) 985448d640bSChris Mason return 0; 986448d640bSChris Mason 987e2d84521SMiao Xie fs_info = BTRFS_I(mapping->host)->root->fs_info; 988b9473439SChris Mason /* this is a bit racy, but that's ok */ 989e2d84521SMiao Xie ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes, 990e2d84521SMiao Xie BTRFS_DIRTY_METADATA_THRESH); 991e2d84521SMiao Xie if (ret < 0) 992793955bcSChris Mason return 0; 993793955bcSChris Mason } 9940b32f4bbSJosef Bacik return btree_write_cache_pages(mapping, wbc); 9950da5468fSChris Mason } 9960da5468fSChris Mason 997b2950863SChristoph Hellwig static int btree_readpage(struct file *file, struct page *page) 9985f39d397SChris Mason { 999d1310b2eSChris Mason struct extent_io_tree *tree; 1000d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 10018ddc7d9cSJan Schmidt return extent_read_full_page(tree, page, btree_get_extent, 0); 10025f39d397SChris Mason } 10035f39d397SChris Mason 100470dec807SChris Mason static int btree_releasepage(struct page *page, gfp_t gfp_flags) 10055f39d397SChris Mason { 100698509cfcSChris Mason if (PageWriteback(page) || PageDirty(page)) 100798509cfcSChris Mason return 0; 10080c4e538bSDavid Sterba 1009f7a52a40SDavid Sterba return try_release_extent_buffer(page); 1010d98237b3SChris Mason } 1011d98237b3SChris Mason 1012d47992f8SLukas Czerner static void btree_invalidatepage(struct page *page, unsigned int offset, 1013d47992f8SLukas Czerner unsigned int length) 1014d98237b3SChris Mason { 1015d1310b2eSChris Mason struct extent_io_tree *tree; 1016d1310b2eSChris Mason tree = &BTRFS_I(page->mapping->host)->io_tree; 10175f39d397SChris Mason extent_invalidatepage(tree, page, offset); 10185f39d397SChris Mason btree_releasepage(page, GFP_NOFS); 10199ad6b7bcSChris Mason if (PagePrivate(page)) { 1020d397712bSChris Mason printk(KERN_WARNING "btrfs warning page private not zero " 1021d397712bSChris Mason "on page %llu\n", (unsigned long long)page_offset(page)); 10229ad6b7bcSChris Mason ClearPagePrivate(page); 10239ad6b7bcSChris Mason set_page_private(page, 0); 10249ad6b7bcSChris Mason page_cache_release(page); 10259ad6b7bcSChris Mason } 1026d98237b3SChris Mason } 1027d98237b3SChris Mason 10280b32f4bbSJosef Bacik static int btree_set_page_dirty(struct page *page) 10290b32f4bbSJosef Bacik { 1030bb146eb2SJosef Bacik #ifdef DEBUG 10310b32f4bbSJosef Bacik struct extent_buffer *eb; 10320b32f4bbSJosef Bacik 10330b32f4bbSJosef Bacik BUG_ON(!PagePrivate(page)); 10340b32f4bbSJosef Bacik eb = (struct extent_buffer *)page->private; 10350b32f4bbSJosef Bacik BUG_ON(!eb); 10360b32f4bbSJosef Bacik BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 10370b32f4bbSJosef Bacik BUG_ON(!atomic_read(&eb->refs)); 10380b32f4bbSJosef Bacik btrfs_assert_tree_locked(eb); 1039bb146eb2SJosef Bacik #endif 10400b32f4bbSJosef Bacik return __set_page_dirty_nobuffers(page); 10410b32f4bbSJosef Bacik } 10420b32f4bbSJosef Bacik 10437f09410bSAlexey Dobriyan static const struct address_space_operations btree_aops = { 1044d98237b3SChris Mason .readpage = btree_readpage, 10450da5468fSChris Mason .writepages = btree_writepages, 10465f39d397SChris Mason .releasepage = btree_releasepage, 10475f39d397SChris Mason .invalidatepage = btree_invalidatepage, 10485a92bc88SChris Mason #ifdef CONFIG_MIGRATION 1049784b4e29SChris Mason .migratepage = btree_migratepage, 10505a92bc88SChris Mason #endif 10510b32f4bbSJosef Bacik .set_page_dirty = btree_set_page_dirty, 1052d98237b3SChris Mason }; 1053123abc88SChris Mason 1054ca7a79adSChris Mason int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize, 1055ca7a79adSChris Mason u64 parent_transid) 1056090d1875SChris Mason { 10575f39d397SChris Mason struct extent_buffer *buf = NULL; 10585f39d397SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 1059de428b63SChris Mason int ret = 0; 1060090d1875SChris Mason 1061db94535dSChris Mason buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 10625f39d397SChris Mason if (!buf) 1063090d1875SChris Mason return 0; 1064d1310b2eSChris Mason read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 1065bb82ab88SArne Jansen buf, 0, WAIT_NONE, btree_get_extent, 0); 10665f39d397SChris Mason free_extent_buffer(buf); 1067de428b63SChris Mason return ret; 1068090d1875SChris Mason } 1069090d1875SChris Mason 1070ab0fff03SArne Jansen int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize, 1071ab0fff03SArne Jansen int mirror_num, struct extent_buffer **eb) 1072ab0fff03SArne Jansen { 1073ab0fff03SArne Jansen struct extent_buffer *buf = NULL; 1074ab0fff03SArne Jansen struct inode *btree_inode = root->fs_info->btree_inode; 1075ab0fff03SArne Jansen struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree; 1076ab0fff03SArne Jansen int ret; 1077ab0fff03SArne Jansen 1078ab0fff03SArne Jansen buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 1079ab0fff03SArne Jansen if (!buf) 1080ab0fff03SArne Jansen return 0; 1081ab0fff03SArne Jansen 1082ab0fff03SArne Jansen set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1083ab0fff03SArne Jansen 1084ab0fff03SArne Jansen ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK, 1085ab0fff03SArne Jansen btree_get_extent, mirror_num); 1086ab0fff03SArne Jansen if (ret) { 1087ab0fff03SArne Jansen free_extent_buffer(buf); 1088ab0fff03SArne Jansen return ret; 1089ab0fff03SArne Jansen } 1090ab0fff03SArne Jansen 1091ab0fff03SArne Jansen if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1092ab0fff03SArne Jansen free_extent_buffer(buf); 1093ab0fff03SArne Jansen return -EIO; 10940b32f4bbSJosef Bacik } else if (extent_buffer_uptodate(buf)) { 1095ab0fff03SArne Jansen *eb = buf; 1096ab0fff03SArne Jansen } else { 1097ab0fff03SArne Jansen free_extent_buffer(buf); 1098ab0fff03SArne Jansen } 1099ab0fff03SArne Jansen return 0; 1100ab0fff03SArne Jansen } 1101ab0fff03SArne Jansen 11020999df54SChris Mason struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root, 11030999df54SChris Mason u64 bytenr, u32 blocksize) 11040999df54SChris Mason { 11050999df54SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 11060999df54SChris Mason struct extent_buffer *eb; 11070999df54SChris Mason eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 1108f09d1f60SDavid Sterba bytenr, blocksize); 11090999df54SChris Mason return eb; 11100999df54SChris Mason } 11110999df54SChris Mason 11120999df54SChris Mason struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root, 11130999df54SChris Mason u64 bytenr, u32 blocksize) 11140999df54SChris Mason { 11150999df54SChris Mason struct inode *btree_inode = root->fs_info->btree_inode; 11160999df54SChris Mason struct extent_buffer *eb; 11170999df54SChris Mason 11180999df54SChris Mason eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree, 1119727011e0SChris Mason bytenr, blocksize); 11200999df54SChris Mason return eb; 11210999df54SChris Mason } 11220999df54SChris Mason 11230999df54SChris Mason 1124e02119d5SChris Mason int btrfs_write_tree_block(struct extent_buffer *buf) 1125e02119d5SChris Mason { 1126727011e0SChris Mason return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, 11278aa38c31SChristoph Hellwig buf->start + buf->len - 1); 1128e02119d5SChris Mason } 1129e02119d5SChris Mason 1130e02119d5SChris Mason int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 1131e02119d5SChris Mason { 1132727011e0SChris Mason return filemap_fdatawait_range(buf->pages[0]->mapping, 11338aa38c31SChristoph Hellwig buf->start, buf->start + buf->len - 1); 1134e02119d5SChris Mason } 1135e02119d5SChris Mason 1136db94535dSChris Mason struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 1137ca7a79adSChris Mason u32 blocksize, u64 parent_transid) 1138e20d96d6SChris Mason { 11395f39d397SChris Mason struct extent_buffer *buf = NULL; 114019c00ddcSChris Mason int ret; 114119c00ddcSChris Mason 1142db94535dSChris Mason buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 11435f39d397SChris Mason if (!buf) 1144d98237b3SChris Mason return NULL; 1145e4204dedSChris Mason 1146ca7a79adSChris Mason ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 11470f0fe8f7SFilipe David Borba Manana if (ret) { 11480f0fe8f7SFilipe David Borba Manana free_extent_buffer(buf); 11490f0fe8f7SFilipe David Borba Manana return NULL; 11500f0fe8f7SFilipe David Borba Manana } 11515f39d397SChris Mason return buf; 1152ce9adaa5SChris Mason 1153eb60ceacSChris Mason } 1154eb60ceacSChris Mason 1155d5c13f92SJeff Mahoney void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, 11565f39d397SChris Mason struct extent_buffer *buf) 1157ed2ff2cbSChris Mason { 1158e2d84521SMiao Xie struct btrfs_fs_info *fs_info = root->fs_info; 1159e2d84521SMiao Xie 116055c69072SChris Mason if (btrfs_header_generation(buf) == 1161e2d84521SMiao Xie fs_info->running_transaction->transid) { 1162b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 1163b4ce94deSChris Mason 1164b9473439SChris Mason if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1165e2d84521SMiao Xie __percpu_counter_add(&fs_info->dirty_metadata_bytes, 1166e2d84521SMiao Xie -buf->len, 1167e2d84521SMiao Xie fs_info->dirty_metadata_batch); 1168b9473439SChris Mason /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1169b9473439SChris Mason btrfs_set_lock_blocking(buf); 11700b32f4bbSJosef Bacik clear_extent_buffer_dirty(buf); 1171925baeddSChris Mason } 11725f39d397SChris Mason } 1173ed7b63ebSJosef Bacik } 11745f39d397SChris Mason 1175143bede5SJeff Mahoney static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize, 117687ee04ebSChris Mason u32 stripesize, struct btrfs_root *root, 11779f5fae2fSChris Mason struct btrfs_fs_info *fs_info, 1178e20d96d6SChris Mason u64 objectid) 1179d97e63b6SChris Mason { 1180cfaa7295SChris Mason root->node = NULL; 1181a28ec197SChris Mason root->commit_root = NULL; 1182db94535dSChris Mason root->sectorsize = sectorsize; 1183db94535dSChris Mason root->nodesize = nodesize; 1184db94535dSChris Mason root->leafsize = leafsize; 118587ee04ebSChris Mason root->stripesize = stripesize; 1186123abc88SChris Mason root->ref_cows = 0; 11870b86a832SChris Mason root->track_dirty = 0; 1188c71bf099SYan, Zheng root->in_radix = 0; 1189d68fc57bSYan, Zheng root->orphan_item_inserted = 0; 1190d68fc57bSYan, Zheng root->orphan_cleanup_state = 0; 11910b86a832SChris Mason 11920f7d52f4SChris Mason root->objectid = objectid; 11930f7d52f4SChris Mason root->last_trans = 0; 119413a8a7c8SYan, Zheng root->highest_objectid = 0; 1195eb73c1b7SMiao Xie root->nr_delalloc_inodes = 0; 1196199c2a9cSMiao Xie root->nr_ordered_extents = 0; 119758176a96SJosef Bacik root->name = NULL; 11986bef4d31SEric Paris root->inode_tree = RB_ROOT; 119916cdcec7SMiao Xie INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1200f0486c68SYan, Zheng root->block_rsv = NULL; 1201d68fc57bSYan, Zheng root->orphan_block_rsv = NULL; 12020b86a832SChris Mason 12030b86a832SChris Mason INIT_LIST_HEAD(&root->dirty_list); 12045d4f98a2SYan Zheng INIT_LIST_HEAD(&root->root_list); 1205eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_inodes); 1206eb73c1b7SMiao Xie INIT_LIST_HEAD(&root->delalloc_root); 1207199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_extents); 1208199c2a9cSMiao Xie INIT_LIST_HEAD(&root->ordered_root); 12092ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[0]); 12102ab28f32SJosef Bacik INIT_LIST_HEAD(&root->logged_list[1]); 1211d68fc57bSYan, Zheng spin_lock_init(&root->orphan_lock); 12125d4f98a2SYan Zheng spin_lock_init(&root->inode_lock); 1213eb73c1b7SMiao Xie spin_lock_init(&root->delalloc_lock); 1214199c2a9cSMiao Xie spin_lock_init(&root->ordered_extent_lock); 1215f0486c68SYan, Zheng spin_lock_init(&root->accounting_lock); 12162ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[0]); 12172ab28f32SJosef Bacik spin_lock_init(&root->log_extents_lock[1]); 1218a2135011SChris Mason mutex_init(&root->objectid_mutex); 1219e02119d5SChris Mason mutex_init(&root->log_mutex); 12207237f183SYan Zheng init_waitqueue_head(&root->log_writer_wait); 12217237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[0]); 12227237f183SYan Zheng init_waitqueue_head(&root->log_commit_wait[1]); 12237237f183SYan Zheng atomic_set(&root->log_commit[0], 0); 12247237f183SYan Zheng atomic_set(&root->log_commit[1], 0); 12257237f183SYan Zheng atomic_set(&root->log_writers, 0); 12262ecb7923SMiao Xie atomic_set(&root->log_batch, 0); 12278a35d95fSJosef Bacik atomic_set(&root->orphan_inodes, 0); 1228b0feb9d9SMiao Xie atomic_set(&root->refs, 1); 12297237f183SYan Zheng root->log_transid = 0; 1230257c62e1SChris Mason root->last_log_commit = 0; 123106ea65a3SJosef Bacik if (fs_info) 1232d0c803c4SChris Mason extent_io_tree_init(&root->dirty_log_pages, 1233f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 1234017e5369SChris Mason 12353768f368SChris Mason memset(&root->root_key, 0, sizeof(root->root_key)); 12363768f368SChris Mason memset(&root->root_item, 0, sizeof(root->root_item)); 12376702ed49SChris Mason memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 123858176a96SJosef Bacik memset(&root->root_kobj, 0, sizeof(root->root_kobj)); 123906ea65a3SJosef Bacik if (fs_info) 12403f157a2fSChris Mason root->defrag_trans_start = fs_info->generation; 124106ea65a3SJosef Bacik else 124206ea65a3SJosef Bacik root->defrag_trans_start = 0; 124358176a96SJosef Bacik init_completion(&root->kobj_unregister); 12446702ed49SChris Mason root->defrag_running = 0; 12454d775673SChris Mason root->root_key.objectid = objectid; 12460ee5dc67SAl Viro root->anon_dev = 0; 12478ea05e3aSAlexander Block 12485f3ab90aSAnand Jain spin_lock_init(&root->root_item_lock); 12493768f368SChris Mason } 12503768f368SChris Mason 1251f84a8bd6SAl Viro static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info) 12526f07e42eSAl Viro { 12536f07e42eSAl Viro struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS); 12546f07e42eSAl Viro if (root) 12556f07e42eSAl Viro root->fs_info = fs_info; 12566f07e42eSAl Viro return root; 12576f07e42eSAl Viro } 12586f07e42eSAl Viro 125906ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 126006ea65a3SJosef Bacik /* Should only be used by the testing infrastructure */ 126106ea65a3SJosef Bacik struct btrfs_root *btrfs_alloc_dummy_root(void) 126206ea65a3SJosef Bacik { 126306ea65a3SJosef Bacik struct btrfs_root *root; 126406ea65a3SJosef Bacik 126506ea65a3SJosef Bacik root = btrfs_alloc_root(NULL); 126606ea65a3SJosef Bacik if (!root) 126706ea65a3SJosef Bacik return ERR_PTR(-ENOMEM); 126806ea65a3SJosef Bacik __setup_root(4096, 4096, 4096, 4096, root, NULL, 1); 126906ea65a3SJosef Bacik root->dummy_root = 1; 127006ea65a3SJosef Bacik 127106ea65a3SJosef Bacik return root; 127206ea65a3SJosef Bacik } 127306ea65a3SJosef Bacik #endif 127406ea65a3SJosef Bacik 127520897f5cSArne Jansen struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 127620897f5cSArne Jansen struct btrfs_fs_info *fs_info, 127720897f5cSArne Jansen u64 objectid) 127820897f5cSArne Jansen { 127920897f5cSArne Jansen struct extent_buffer *leaf; 128020897f5cSArne Jansen struct btrfs_root *tree_root = fs_info->tree_root; 128120897f5cSArne Jansen struct btrfs_root *root; 128220897f5cSArne Jansen struct btrfs_key key; 128320897f5cSArne Jansen int ret = 0; 128420897f5cSArne Jansen u64 bytenr; 12856463fe58SStefan Behrens uuid_le uuid; 128620897f5cSArne Jansen 128720897f5cSArne Jansen root = btrfs_alloc_root(fs_info); 128820897f5cSArne Jansen if (!root) 128920897f5cSArne Jansen return ERR_PTR(-ENOMEM); 129020897f5cSArne Jansen 129120897f5cSArne Jansen __setup_root(tree_root->nodesize, tree_root->leafsize, 129220897f5cSArne Jansen tree_root->sectorsize, tree_root->stripesize, 129320897f5cSArne Jansen root, fs_info, objectid); 129420897f5cSArne Jansen root->root_key.objectid = objectid; 129520897f5cSArne Jansen root->root_key.type = BTRFS_ROOT_ITEM_KEY; 129620897f5cSArne Jansen root->root_key.offset = 0; 129720897f5cSArne Jansen 129820897f5cSArne Jansen leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 129920897f5cSArne Jansen 0, objectid, NULL, 0, 0, 0); 130020897f5cSArne Jansen if (IS_ERR(leaf)) { 130120897f5cSArne Jansen ret = PTR_ERR(leaf); 13021dd05682STsutomu Itoh leaf = NULL; 130320897f5cSArne Jansen goto fail; 130420897f5cSArne Jansen } 130520897f5cSArne Jansen 130620897f5cSArne Jansen bytenr = leaf->start; 130720897f5cSArne Jansen memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 130820897f5cSArne Jansen btrfs_set_header_bytenr(leaf, leaf->start); 130920897f5cSArne Jansen btrfs_set_header_generation(leaf, trans->transid); 131020897f5cSArne Jansen btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 131120897f5cSArne Jansen btrfs_set_header_owner(leaf, objectid); 131220897f5cSArne Jansen root->node = leaf; 131320897f5cSArne Jansen 13140a4e5586SRoss Kirk write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(), 131520897f5cSArne Jansen BTRFS_FSID_SIZE); 131620897f5cSArne Jansen write_extent_buffer(leaf, fs_info->chunk_tree_uuid, 1317b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(leaf), 131820897f5cSArne Jansen BTRFS_UUID_SIZE); 131920897f5cSArne Jansen btrfs_mark_buffer_dirty(leaf); 132020897f5cSArne Jansen 132120897f5cSArne Jansen root->commit_root = btrfs_root_node(root); 132220897f5cSArne Jansen root->track_dirty = 1; 132320897f5cSArne Jansen 132420897f5cSArne Jansen 132520897f5cSArne Jansen root->root_item.flags = 0; 132620897f5cSArne Jansen root->root_item.byte_limit = 0; 132720897f5cSArne Jansen btrfs_set_root_bytenr(&root->root_item, leaf->start); 132820897f5cSArne Jansen btrfs_set_root_generation(&root->root_item, trans->transid); 132920897f5cSArne Jansen btrfs_set_root_level(&root->root_item, 0); 133020897f5cSArne Jansen btrfs_set_root_refs(&root->root_item, 1); 133120897f5cSArne Jansen btrfs_set_root_used(&root->root_item, leaf->len); 133220897f5cSArne Jansen btrfs_set_root_last_snapshot(&root->root_item, 0); 133320897f5cSArne Jansen btrfs_set_root_dirid(&root->root_item, 0); 13346463fe58SStefan Behrens uuid_le_gen(&uuid); 13356463fe58SStefan Behrens memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 133620897f5cSArne Jansen root->root_item.drop_level = 0; 133720897f5cSArne Jansen 133820897f5cSArne Jansen key.objectid = objectid; 133920897f5cSArne Jansen key.type = BTRFS_ROOT_ITEM_KEY; 134020897f5cSArne Jansen key.offset = 0; 134120897f5cSArne Jansen ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 134220897f5cSArne Jansen if (ret) 134320897f5cSArne Jansen goto fail; 134420897f5cSArne Jansen 134520897f5cSArne Jansen btrfs_tree_unlock(leaf); 134620897f5cSArne Jansen 134720897f5cSArne Jansen return root; 13481dd05682STsutomu Itoh 13491dd05682STsutomu Itoh fail: 13501dd05682STsutomu Itoh if (leaf) { 13511dd05682STsutomu Itoh btrfs_tree_unlock(leaf); 13521dd05682STsutomu Itoh free_extent_buffer(leaf); 13531dd05682STsutomu Itoh } 13541dd05682STsutomu Itoh kfree(root); 13551dd05682STsutomu Itoh 13561dd05682STsutomu Itoh return ERR_PTR(ret); 135720897f5cSArne Jansen } 135820897f5cSArne Jansen 13597237f183SYan Zheng static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1360e02119d5SChris Mason struct btrfs_fs_info *fs_info) 13610f7d52f4SChris Mason { 13620f7d52f4SChris Mason struct btrfs_root *root; 13630f7d52f4SChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 13647237f183SYan Zheng struct extent_buffer *leaf; 1365e02119d5SChris Mason 13666f07e42eSAl Viro root = btrfs_alloc_root(fs_info); 1367e02119d5SChris Mason if (!root) 13687237f183SYan Zheng return ERR_PTR(-ENOMEM); 1369e02119d5SChris Mason 1370e02119d5SChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 1371e02119d5SChris Mason tree_root->sectorsize, tree_root->stripesize, 1372e02119d5SChris Mason root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1373e02119d5SChris Mason 1374e02119d5SChris Mason root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1375e02119d5SChris Mason root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1376e02119d5SChris Mason root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 13777237f183SYan Zheng /* 13787237f183SYan Zheng * log trees do not get reference counted because they go away 13797237f183SYan Zheng * before a real commit is actually done. They do store pointers 13807237f183SYan Zheng * to file data extents, and those reference counts still get 13817237f183SYan Zheng * updated (along with back refs to the log tree). 13827237f183SYan Zheng */ 1383e02119d5SChris Mason root->ref_cows = 0; 1384e02119d5SChris Mason 13855d4f98a2SYan Zheng leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, 138666d7e7f0SArne Jansen BTRFS_TREE_LOG_OBJECTID, NULL, 13875581a51aSJan Schmidt 0, 0, 0); 13887237f183SYan Zheng if (IS_ERR(leaf)) { 13897237f183SYan Zheng kfree(root); 13907237f183SYan Zheng return ERR_CAST(leaf); 13917237f183SYan Zheng } 1392e02119d5SChris Mason 13935d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 13945d4f98a2SYan Zheng btrfs_set_header_bytenr(leaf, leaf->start); 13955d4f98a2SYan Zheng btrfs_set_header_generation(leaf, trans->transid); 13965d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 13975d4f98a2SYan Zheng btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 13987237f183SYan Zheng root->node = leaf; 1399e02119d5SChris Mason 1400e02119d5SChris Mason write_extent_buffer(root->node, root->fs_info->fsid, 14010a4e5586SRoss Kirk btrfs_header_fsid(), BTRFS_FSID_SIZE); 1402e02119d5SChris Mason btrfs_mark_buffer_dirty(root->node); 1403e02119d5SChris Mason btrfs_tree_unlock(root->node); 14047237f183SYan Zheng return root; 14057237f183SYan Zheng } 14067237f183SYan Zheng 14077237f183SYan Zheng int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 14087237f183SYan Zheng struct btrfs_fs_info *fs_info) 14097237f183SYan Zheng { 14107237f183SYan Zheng struct btrfs_root *log_root; 14117237f183SYan Zheng 14127237f183SYan Zheng log_root = alloc_log_tree(trans, fs_info); 14137237f183SYan Zheng if (IS_ERR(log_root)) 14147237f183SYan Zheng return PTR_ERR(log_root); 14157237f183SYan Zheng WARN_ON(fs_info->log_root_tree); 14167237f183SYan Zheng fs_info->log_root_tree = log_root; 14177237f183SYan Zheng return 0; 14187237f183SYan Zheng } 14197237f183SYan Zheng 14207237f183SYan Zheng int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 14217237f183SYan Zheng struct btrfs_root *root) 14227237f183SYan Zheng { 14237237f183SYan Zheng struct btrfs_root *log_root; 14247237f183SYan Zheng struct btrfs_inode_item *inode_item; 14257237f183SYan Zheng 14267237f183SYan Zheng log_root = alloc_log_tree(trans, root->fs_info); 14277237f183SYan Zheng if (IS_ERR(log_root)) 14287237f183SYan Zheng return PTR_ERR(log_root); 14297237f183SYan Zheng 14307237f183SYan Zheng log_root->last_trans = trans->transid; 14317237f183SYan Zheng log_root->root_key.offset = root->root_key.objectid; 14327237f183SYan Zheng 14337237f183SYan Zheng inode_item = &log_root->root_item.inode; 14343cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 14353cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 14363cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 14373cae210fSQu Wenruo btrfs_set_stack_inode_nbytes(inode_item, root->leafsize); 14383cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 14397237f183SYan Zheng 14405d4f98a2SYan Zheng btrfs_set_root_node(&log_root->root_item, log_root->node); 14417237f183SYan Zheng 14427237f183SYan Zheng WARN_ON(root->log_root); 14437237f183SYan Zheng root->log_root = log_root; 14447237f183SYan Zheng root->log_transid = 0; 1445257c62e1SChris Mason root->last_log_commit = 0; 1446e02119d5SChris Mason return 0; 1447e02119d5SChris Mason } 1448e02119d5SChris Mason 144935a3621bSStefan Behrens static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1450cb517eabSMiao Xie struct btrfs_key *key) 1451e02119d5SChris Mason { 1452e02119d5SChris Mason struct btrfs_root *root; 1453e02119d5SChris Mason struct btrfs_fs_info *fs_info = tree_root->fs_info; 14540f7d52f4SChris Mason struct btrfs_path *path; 145584234f3aSYan Zheng u64 generation; 1456db94535dSChris Mason u32 blocksize; 1457cb517eabSMiao Xie int ret; 1458cb517eabSMiao Xie 1459cb517eabSMiao Xie path = btrfs_alloc_path(); 1460cb517eabSMiao Xie if (!path) 1461cb517eabSMiao Xie return ERR_PTR(-ENOMEM); 14620f7d52f4SChris Mason 14636f07e42eSAl Viro root = btrfs_alloc_root(fs_info); 1464cb517eabSMiao Xie if (!root) { 1465cb517eabSMiao Xie ret = -ENOMEM; 1466cb517eabSMiao Xie goto alloc_fail; 14670f7d52f4SChris Mason } 14680f7d52f4SChris Mason 1469db94535dSChris Mason __setup_root(tree_root->nodesize, tree_root->leafsize, 147087ee04ebSChris Mason tree_root->sectorsize, tree_root->stripesize, 1471cb517eabSMiao Xie root, fs_info, key->objectid); 14720f7d52f4SChris Mason 1473cb517eabSMiao Xie ret = btrfs_find_root(tree_root, key, path, 1474cb517eabSMiao Xie &root->root_item, &root->root_key); 14750f7d52f4SChris Mason if (ret) { 147613a8a7c8SYan, Zheng if (ret > 0) 147713a8a7c8SYan, Zheng ret = -ENOENT; 1478cb517eabSMiao Xie goto find_fail; 14790f7d52f4SChris Mason } 148013a8a7c8SYan, Zheng 148184234f3aSYan Zheng generation = btrfs_root_generation(&root->root_item); 1482db94535dSChris Mason blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1483db94535dSChris Mason root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 148484234f3aSYan Zheng blocksize, generation); 1485cb517eabSMiao Xie if (!root->node) { 1486cb517eabSMiao Xie ret = -ENOMEM; 1487cb517eabSMiao Xie goto find_fail; 1488cb517eabSMiao Xie } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1489cb517eabSMiao Xie ret = -EIO; 1490cb517eabSMiao Xie goto read_fail; 1491416bc658SJosef Bacik } 14925d4f98a2SYan Zheng root->commit_root = btrfs_root_node(root); 149313a8a7c8SYan, Zheng out: 1494cb517eabSMiao Xie btrfs_free_path(path); 1495cb517eabSMiao Xie return root; 1496cb517eabSMiao Xie 1497cb517eabSMiao Xie read_fail: 1498cb517eabSMiao Xie free_extent_buffer(root->node); 1499cb517eabSMiao Xie find_fail: 1500cb517eabSMiao Xie kfree(root); 1501cb517eabSMiao Xie alloc_fail: 1502cb517eabSMiao Xie root = ERR_PTR(ret); 1503cb517eabSMiao Xie goto out; 1504cb517eabSMiao Xie } 1505cb517eabSMiao Xie 1506cb517eabSMiao Xie struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1507cb517eabSMiao Xie struct btrfs_key *location) 1508cb517eabSMiao Xie { 1509cb517eabSMiao Xie struct btrfs_root *root; 1510cb517eabSMiao Xie 1511cb517eabSMiao Xie root = btrfs_read_tree_root(tree_root, location); 1512cb517eabSMiao Xie if (IS_ERR(root)) 1513cb517eabSMiao Xie return root; 1514cb517eabSMiao Xie 1515cb517eabSMiao Xie if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 15160f7d52f4SChris Mason root->ref_cows = 1; 151708fe4db1SLi Zefan btrfs_check_and_init_root_item(&root->root_item); 151808fe4db1SLi Zefan } 151913a8a7c8SYan, Zheng 15205eda7b5eSChris Mason return root; 15215eda7b5eSChris Mason } 15225eda7b5eSChris Mason 1523cb517eabSMiao Xie int btrfs_init_fs_root(struct btrfs_root *root) 1524cb517eabSMiao Xie { 1525cb517eabSMiao Xie int ret; 1526cb517eabSMiao Xie 1527cb517eabSMiao Xie root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1528cb517eabSMiao Xie root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1529cb517eabSMiao Xie GFP_NOFS); 1530cb517eabSMiao Xie if (!root->free_ino_pinned || !root->free_ino_ctl) { 1531cb517eabSMiao Xie ret = -ENOMEM; 1532cb517eabSMiao Xie goto fail; 1533cb517eabSMiao Xie } 1534cb517eabSMiao Xie 1535cb517eabSMiao Xie btrfs_init_free_ino_ctl(root); 1536cb517eabSMiao Xie mutex_init(&root->fs_commit_mutex); 1537cb517eabSMiao Xie spin_lock_init(&root->cache_lock); 1538cb517eabSMiao Xie init_waitqueue_head(&root->cache_wait); 1539cb517eabSMiao Xie 1540cb517eabSMiao Xie ret = get_anon_bdev(&root->anon_dev); 1541cb517eabSMiao Xie if (ret) 1542cb517eabSMiao Xie goto fail; 1543cb517eabSMiao Xie return 0; 1544cb517eabSMiao Xie fail: 1545cb517eabSMiao Xie kfree(root->free_ino_ctl); 1546cb517eabSMiao Xie kfree(root->free_ino_pinned); 1547cb517eabSMiao Xie return ret; 1548cb517eabSMiao Xie } 1549cb517eabSMiao Xie 1550171170c1SSergei Trofimovich static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1551cb517eabSMiao Xie u64 root_id) 1552cb517eabSMiao Xie { 1553cb517eabSMiao Xie struct btrfs_root *root; 1554cb517eabSMiao Xie 1555cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1556cb517eabSMiao Xie root = radix_tree_lookup(&fs_info->fs_roots_radix, 1557cb517eabSMiao Xie (unsigned long)root_id); 1558cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1559cb517eabSMiao Xie return root; 1560cb517eabSMiao Xie } 1561cb517eabSMiao Xie 1562cb517eabSMiao Xie int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1563cb517eabSMiao Xie struct btrfs_root *root) 1564cb517eabSMiao Xie { 1565cb517eabSMiao Xie int ret; 1566cb517eabSMiao Xie 1567cb517eabSMiao Xie ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 1568cb517eabSMiao Xie if (ret) 1569cb517eabSMiao Xie return ret; 1570cb517eabSMiao Xie 1571cb517eabSMiao Xie spin_lock(&fs_info->fs_roots_radix_lock); 1572cb517eabSMiao Xie ret = radix_tree_insert(&fs_info->fs_roots_radix, 1573cb517eabSMiao Xie (unsigned long)root->root_key.objectid, 1574cb517eabSMiao Xie root); 1575cb517eabSMiao Xie if (ret == 0) 1576cb517eabSMiao Xie root->in_radix = 1; 1577cb517eabSMiao Xie spin_unlock(&fs_info->fs_roots_radix_lock); 1578cb517eabSMiao Xie radix_tree_preload_end(); 1579cb517eabSMiao Xie 1580cb517eabSMiao Xie return ret; 1581cb517eabSMiao Xie } 1582cb517eabSMiao Xie 1583c00869f1SMiao Xie struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1584c00869f1SMiao Xie struct btrfs_key *location, 1585c00869f1SMiao Xie bool check_ref) 15865eda7b5eSChris Mason { 15875eda7b5eSChris Mason struct btrfs_root *root; 15885eda7b5eSChris Mason int ret; 15895eda7b5eSChris Mason 1590edbd8d4eSChris Mason if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1591edbd8d4eSChris Mason return fs_info->tree_root; 1592edbd8d4eSChris Mason if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1593edbd8d4eSChris Mason return fs_info->extent_root; 15948f18cf13SChris Mason if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 15958f18cf13SChris Mason return fs_info->chunk_root; 15968f18cf13SChris Mason if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 15978f18cf13SChris Mason return fs_info->dev_root; 15980403e47eSYan Zheng if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 15990403e47eSYan Zheng return fs_info->csum_root; 1600bcef60f2SArne Jansen if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1601bcef60f2SArne Jansen return fs_info->quota_root ? fs_info->quota_root : 1602bcef60f2SArne Jansen ERR_PTR(-ENOENT); 1603f7a81ea4SStefan Behrens if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1604f7a81ea4SStefan Behrens return fs_info->uuid_root ? fs_info->uuid_root : 1605f7a81ea4SStefan Behrens ERR_PTR(-ENOENT); 16064df27c4dSYan, Zheng again: 1607cb517eabSMiao Xie root = btrfs_lookup_fs_root(fs_info, location->objectid); 160848475471SStefan Behrens if (root) { 1609c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) 161048475471SStefan Behrens return ERR_PTR(-ENOENT); 16115eda7b5eSChris Mason return root; 161248475471SStefan Behrens } 16135eda7b5eSChris Mason 1614cb517eabSMiao Xie root = btrfs_read_fs_root(fs_info->tree_root, location); 16155eda7b5eSChris Mason if (IS_ERR(root)) 16165eda7b5eSChris Mason return root; 16173394e160SChris Mason 1618c00869f1SMiao Xie if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1619d68fc57bSYan, Zheng ret = -ENOENT; 1620d68fc57bSYan, Zheng goto fail; 1621d68fc57bSYan, Zheng } 1622d68fc57bSYan, Zheng 1623cb517eabSMiao Xie ret = btrfs_init_fs_root(root); 1624cb517eabSMiao Xie if (ret) 1625cb517eabSMiao Xie goto fail; 1626cb517eabSMiao Xie 1627d68fc57bSYan, Zheng ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid); 1628d68fc57bSYan, Zheng if (ret < 0) 1629d68fc57bSYan, Zheng goto fail; 1630d68fc57bSYan, Zheng if (ret == 0) 1631d68fc57bSYan, Zheng root->orphan_item_inserted = 1; 1632d68fc57bSYan, Zheng 1633cb517eabSMiao Xie ret = btrfs_insert_fs_root(fs_info, root); 16340f7d52f4SChris Mason if (ret) { 16354df27c4dSYan, Zheng if (ret == -EEXIST) { 16364df27c4dSYan, Zheng free_fs_root(root); 16374df27c4dSYan, Zheng goto again; 16380f7d52f4SChris Mason } 16394df27c4dSYan, Zheng goto fail; 16404df27c4dSYan, Zheng } 1641edbd8d4eSChris Mason return root; 16424df27c4dSYan, Zheng fail: 16434df27c4dSYan, Zheng free_fs_root(root); 16444df27c4dSYan, Zheng return ERR_PTR(ret); 1645edbd8d4eSChris Mason } 1646edbd8d4eSChris Mason 164704160088SChris Mason static int btrfs_congested_fn(void *congested_data, int bdi_bits) 164804160088SChris Mason { 164904160088SChris Mason struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 165004160088SChris Mason int ret = 0; 165104160088SChris Mason struct btrfs_device *device; 165204160088SChris Mason struct backing_dev_info *bdi; 1653b7967db7SChris Mason 16541f78160cSXiao Guangrong rcu_read_lock(); 16551f78160cSXiao Guangrong list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1656dfe25020SChris Mason if (!device->bdev) 1657dfe25020SChris Mason continue; 165804160088SChris Mason bdi = blk_get_backing_dev_info(device->bdev); 165904160088SChris Mason if (bdi && bdi_congested(bdi, bdi_bits)) { 166004160088SChris Mason ret = 1; 166104160088SChris Mason break; 166204160088SChris Mason } 166304160088SChris Mason } 16641f78160cSXiao Guangrong rcu_read_unlock(); 166504160088SChris Mason return ret; 166604160088SChris Mason } 166704160088SChris Mason 166838b66988SChris Mason /* 1669ad081f14SJens Axboe * If this fails, caller must call bdi_destroy() to get rid of the 1670ad081f14SJens Axboe * bdi again. 1671ad081f14SJens Axboe */ 167204160088SChris Mason static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi) 167304160088SChris Mason { 1674ad081f14SJens Axboe int err; 1675ad081f14SJens Axboe 1676ad081f14SJens Axboe bdi->capabilities = BDI_CAP_MAP_COPY; 1677e6d086d8SJens Axboe err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY); 1678ad081f14SJens Axboe if (err) 1679ad081f14SJens Axboe return err; 1680ad081f14SJens Axboe 16814575c9ccSChris Mason bdi->ra_pages = default_backing_dev_info.ra_pages; 168204160088SChris Mason bdi->congested_fn = btrfs_congested_fn; 168304160088SChris Mason bdi->congested_data = info; 168404160088SChris Mason return 0; 168504160088SChris Mason } 168604160088SChris Mason 16878b712842SChris Mason /* 16888b712842SChris Mason * called by the kthread helper functions to finally call the bio end_io 16898b712842SChris Mason * functions. This is where read checksum verification actually happens 16908b712842SChris Mason */ 16918b712842SChris Mason static void end_workqueue_fn(struct btrfs_work *work) 1692ce9adaa5SChris Mason { 1693ce9adaa5SChris Mason struct bio *bio; 16948b712842SChris Mason struct end_io_wq *end_io_wq; 16958b712842SChris Mason struct btrfs_fs_info *fs_info; 1696ce9adaa5SChris Mason int error; 1697ce9adaa5SChris Mason 16988b712842SChris Mason end_io_wq = container_of(work, struct end_io_wq, work); 1699ce9adaa5SChris Mason bio = end_io_wq->bio; 17008b712842SChris Mason fs_info = end_io_wq->info; 17018b712842SChris Mason 1702ce9adaa5SChris Mason error = end_io_wq->error; 1703ce9adaa5SChris Mason bio->bi_private = end_io_wq->private; 1704ce9adaa5SChris Mason bio->bi_end_io = end_io_wq->end_io; 1705ce9adaa5SChris Mason kfree(end_io_wq); 1706ce9adaa5SChris Mason bio_endio(bio, error); 1707ce9adaa5SChris Mason } 170844b8bd7eSChris Mason 1709a74a4b97SChris Mason static int cleaner_kthread(void *arg) 1710a74a4b97SChris Mason { 1711a74a4b97SChris Mason struct btrfs_root *root = arg; 1712d0278245SMiao Xie int again; 1713a74a4b97SChris Mason 1714a74a4b97SChris Mason do { 1715d0278245SMiao Xie again = 0; 17169d1a2a3aSDavid Sterba 1717d0278245SMiao Xie /* Make the cleaner go to sleep early. */ 1718babbf170SMiao Xie if (btrfs_need_cleaner_sleep(root)) 1719d0278245SMiao Xie goto sleep; 1720d0278245SMiao Xie 1721d0278245SMiao Xie if (!mutex_trylock(&root->fs_info->cleaner_mutex)) 1722d0278245SMiao Xie goto sleep; 1723d0278245SMiao Xie 1724dc7f370cSMiao Xie /* 1725dc7f370cSMiao Xie * Avoid the problem that we change the status of the fs 1726dc7f370cSMiao Xie * during the above check and trylock. 1727dc7f370cSMiao Xie */ 1728babbf170SMiao Xie if (btrfs_need_cleaner_sleep(root)) { 1729dc7f370cSMiao Xie mutex_unlock(&root->fs_info->cleaner_mutex); 1730dc7f370cSMiao Xie goto sleep; 1731dc7f370cSMiao Xie } 1732dc7f370cSMiao Xie 173324bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 17349d1a2a3aSDavid Sterba again = btrfs_clean_one_deleted_snapshot(root); 1735a74a4b97SChris Mason mutex_unlock(&root->fs_info->cleaner_mutex); 1736a74a4b97SChris Mason 1737d0278245SMiao Xie /* 173805323cd1SMiao Xie * The defragger has dealt with the R/O remount and umount, 173905323cd1SMiao Xie * needn't do anything special here. 1740d0278245SMiao Xie */ 1741d0278245SMiao Xie btrfs_run_defrag_inodes(root->fs_info); 1742d0278245SMiao Xie sleep: 17439d1a2a3aSDavid Sterba if (!try_to_freeze() && !again) { 1744a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 17458929ecfaSYan, Zheng if (!kthread_should_stop()) 1746a74a4b97SChris Mason schedule(); 1747a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1748a74a4b97SChris Mason } 1749a74a4b97SChris Mason } while (!kthread_should_stop()); 1750a74a4b97SChris Mason return 0; 1751a74a4b97SChris Mason } 1752a74a4b97SChris Mason 1753a74a4b97SChris Mason static int transaction_kthread(void *arg) 1754a74a4b97SChris Mason { 1755a74a4b97SChris Mason struct btrfs_root *root = arg; 1756a74a4b97SChris Mason struct btrfs_trans_handle *trans; 1757a74a4b97SChris Mason struct btrfs_transaction *cur; 17588929ecfaSYan, Zheng u64 transid; 1759a74a4b97SChris Mason unsigned long now; 1760a74a4b97SChris Mason unsigned long delay; 1761914b2007SJan Kara bool cannot_commit; 1762a74a4b97SChris Mason 1763a74a4b97SChris Mason do { 1764914b2007SJan Kara cannot_commit = false; 17658b87dc17SDavid Sterba delay = HZ * root->fs_info->commit_interval; 1766a74a4b97SChris Mason mutex_lock(&root->fs_info->transaction_kthread_mutex); 1767a74a4b97SChris Mason 1768a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1769a74a4b97SChris Mason cur = root->fs_info->running_transaction; 1770a74a4b97SChris Mason if (!cur) { 1771a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1772a74a4b97SChris Mason goto sleep; 1773a74a4b97SChris Mason } 177431153d81SYan Zheng 1775a74a4b97SChris Mason now = get_seconds(); 17764a9d8bdeSMiao Xie if (cur->state < TRANS_STATE_BLOCKED && 17778b87dc17SDavid Sterba (now < cur->start_time || 17788b87dc17SDavid Sterba now - cur->start_time < root->fs_info->commit_interval)) { 1779a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1780a74a4b97SChris Mason delay = HZ * 5; 1781a74a4b97SChris Mason goto sleep; 1782a74a4b97SChris Mason } 17838929ecfaSYan, Zheng transid = cur->transid; 1784a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 178556bec294SChris Mason 178679787eaaSJeff Mahoney /* If the file system is aborted, this will always fail. */ 1787354aa0fbSMiao Xie trans = btrfs_attach_transaction(root); 1788914b2007SJan Kara if (IS_ERR(trans)) { 1789354aa0fbSMiao Xie if (PTR_ERR(trans) != -ENOENT) 1790914b2007SJan Kara cannot_commit = true; 179179787eaaSJeff Mahoney goto sleep; 1792914b2007SJan Kara } 17938929ecfaSYan, Zheng if (transid == trans->transid) { 179479787eaaSJeff Mahoney btrfs_commit_transaction(trans, root); 17958929ecfaSYan, Zheng } else { 17968929ecfaSYan, Zheng btrfs_end_transaction(trans, root); 17978929ecfaSYan, Zheng } 1798a74a4b97SChris Mason sleep: 1799a74a4b97SChris Mason wake_up_process(root->fs_info->cleaner_kthread); 1800a74a4b97SChris Mason mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1801a74a4b97SChris Mason 18024e121c06SJosef Bacik if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 18034e121c06SJosef Bacik &root->fs_info->fs_state))) 18044e121c06SJosef Bacik btrfs_cleanup_transaction(root); 1805a0acae0eSTejun Heo if (!try_to_freeze()) { 1806a74a4b97SChris Mason set_current_state(TASK_INTERRUPTIBLE); 18078929ecfaSYan, Zheng if (!kthread_should_stop() && 1808914b2007SJan Kara (!btrfs_transaction_blocked(root->fs_info) || 1809914b2007SJan Kara cannot_commit)) 1810a74a4b97SChris Mason schedule_timeout(delay); 1811a74a4b97SChris Mason __set_current_state(TASK_RUNNING); 1812a74a4b97SChris Mason } 1813a74a4b97SChris Mason } while (!kthread_should_stop()); 1814a74a4b97SChris Mason return 0; 1815a74a4b97SChris Mason } 1816a74a4b97SChris Mason 1817af31f5e5SChris Mason /* 1818af31f5e5SChris Mason * this will find the highest generation in the array of 1819af31f5e5SChris Mason * root backups. The index of the highest array is returned, 1820af31f5e5SChris Mason * or -1 if we can't find anything. 1821af31f5e5SChris Mason * 1822af31f5e5SChris Mason * We check to make sure the array is valid by comparing the 1823af31f5e5SChris Mason * generation of the latest root in the array with the generation 1824af31f5e5SChris Mason * in the super block. If they don't match we pitch it. 1825af31f5e5SChris Mason */ 1826af31f5e5SChris Mason static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1827af31f5e5SChris Mason { 1828af31f5e5SChris Mason u64 cur; 1829af31f5e5SChris Mason int newest_index = -1; 1830af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1831af31f5e5SChris Mason int i; 1832af31f5e5SChris Mason 1833af31f5e5SChris Mason for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1834af31f5e5SChris Mason root_backup = info->super_copy->super_roots + i; 1835af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1836af31f5e5SChris Mason if (cur == newest_gen) 1837af31f5e5SChris Mason newest_index = i; 1838af31f5e5SChris Mason } 1839af31f5e5SChris Mason 1840af31f5e5SChris Mason /* check to see if we actually wrapped around */ 1841af31f5e5SChris Mason if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1842af31f5e5SChris Mason root_backup = info->super_copy->super_roots; 1843af31f5e5SChris Mason cur = btrfs_backup_tree_root_gen(root_backup); 1844af31f5e5SChris Mason if (cur == newest_gen) 1845af31f5e5SChris Mason newest_index = 0; 1846af31f5e5SChris Mason } 1847af31f5e5SChris Mason return newest_index; 1848af31f5e5SChris Mason } 1849af31f5e5SChris Mason 1850af31f5e5SChris Mason 1851af31f5e5SChris Mason /* 1852af31f5e5SChris Mason * find the oldest backup so we know where to store new entries 1853af31f5e5SChris Mason * in the backup array. This will set the backup_root_index 1854af31f5e5SChris Mason * field in the fs_info struct 1855af31f5e5SChris Mason */ 1856af31f5e5SChris Mason static void find_oldest_super_backup(struct btrfs_fs_info *info, 1857af31f5e5SChris Mason u64 newest_gen) 1858af31f5e5SChris Mason { 1859af31f5e5SChris Mason int newest_index = -1; 1860af31f5e5SChris Mason 1861af31f5e5SChris Mason newest_index = find_newest_super_backup(info, newest_gen); 1862af31f5e5SChris Mason /* if there was garbage in there, just move along */ 1863af31f5e5SChris Mason if (newest_index == -1) { 1864af31f5e5SChris Mason info->backup_root_index = 0; 1865af31f5e5SChris Mason } else { 1866af31f5e5SChris Mason info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1867af31f5e5SChris Mason } 1868af31f5e5SChris Mason } 1869af31f5e5SChris Mason 1870af31f5e5SChris Mason /* 1871af31f5e5SChris Mason * copy all the root pointers into the super backup array. 1872af31f5e5SChris Mason * this will bump the backup pointer by one when it is 1873af31f5e5SChris Mason * done 1874af31f5e5SChris Mason */ 1875af31f5e5SChris Mason static void backup_super_roots(struct btrfs_fs_info *info) 1876af31f5e5SChris Mason { 1877af31f5e5SChris Mason int next_backup; 1878af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1879af31f5e5SChris Mason int last_backup; 1880af31f5e5SChris Mason 1881af31f5e5SChris Mason next_backup = info->backup_root_index; 1882af31f5e5SChris Mason last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1883af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1884af31f5e5SChris Mason 1885af31f5e5SChris Mason /* 1886af31f5e5SChris Mason * just overwrite the last backup if we're at the same generation 1887af31f5e5SChris Mason * this happens only at umount 1888af31f5e5SChris Mason */ 1889af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + last_backup; 1890af31f5e5SChris Mason if (btrfs_backup_tree_root_gen(root_backup) == 1891af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)) 1892af31f5e5SChris Mason next_backup = last_backup; 1893af31f5e5SChris Mason 1894af31f5e5SChris Mason root_backup = info->super_for_commit->super_roots + next_backup; 1895af31f5e5SChris Mason 1896af31f5e5SChris Mason /* 1897af31f5e5SChris Mason * make sure all of our padding and empty slots get zero filled 1898af31f5e5SChris Mason * regardless of which ones we use today 1899af31f5e5SChris Mason */ 1900af31f5e5SChris Mason memset(root_backup, 0, sizeof(*root_backup)); 1901af31f5e5SChris Mason 1902af31f5e5SChris Mason info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1903af31f5e5SChris Mason 1904af31f5e5SChris Mason btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1905af31f5e5SChris Mason btrfs_set_backup_tree_root_gen(root_backup, 1906af31f5e5SChris Mason btrfs_header_generation(info->tree_root->node)); 1907af31f5e5SChris Mason 1908af31f5e5SChris Mason btrfs_set_backup_tree_root_level(root_backup, 1909af31f5e5SChris Mason btrfs_header_level(info->tree_root->node)); 1910af31f5e5SChris Mason 1911af31f5e5SChris Mason btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1912af31f5e5SChris Mason btrfs_set_backup_chunk_root_gen(root_backup, 1913af31f5e5SChris Mason btrfs_header_generation(info->chunk_root->node)); 1914af31f5e5SChris Mason btrfs_set_backup_chunk_root_level(root_backup, 1915af31f5e5SChris Mason btrfs_header_level(info->chunk_root->node)); 1916af31f5e5SChris Mason 1917af31f5e5SChris Mason btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1918af31f5e5SChris Mason btrfs_set_backup_extent_root_gen(root_backup, 1919af31f5e5SChris Mason btrfs_header_generation(info->extent_root->node)); 1920af31f5e5SChris Mason btrfs_set_backup_extent_root_level(root_backup, 1921af31f5e5SChris Mason btrfs_header_level(info->extent_root->node)); 1922af31f5e5SChris Mason 19237c7e82a7SChris Mason /* 19247c7e82a7SChris Mason * we might commit during log recovery, which happens before we set 19257c7e82a7SChris Mason * the fs_root. Make sure it is valid before we fill it in. 19267c7e82a7SChris Mason */ 19277c7e82a7SChris Mason if (info->fs_root && info->fs_root->node) { 19287c7e82a7SChris Mason btrfs_set_backup_fs_root(root_backup, 19297c7e82a7SChris Mason info->fs_root->node->start); 1930af31f5e5SChris Mason btrfs_set_backup_fs_root_gen(root_backup, 1931af31f5e5SChris Mason btrfs_header_generation(info->fs_root->node)); 1932af31f5e5SChris Mason btrfs_set_backup_fs_root_level(root_backup, 1933af31f5e5SChris Mason btrfs_header_level(info->fs_root->node)); 19347c7e82a7SChris Mason } 1935af31f5e5SChris Mason 1936af31f5e5SChris Mason btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1937af31f5e5SChris Mason btrfs_set_backup_dev_root_gen(root_backup, 1938af31f5e5SChris Mason btrfs_header_generation(info->dev_root->node)); 1939af31f5e5SChris Mason btrfs_set_backup_dev_root_level(root_backup, 1940af31f5e5SChris Mason btrfs_header_level(info->dev_root->node)); 1941af31f5e5SChris Mason 1942af31f5e5SChris Mason btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1943af31f5e5SChris Mason btrfs_set_backup_csum_root_gen(root_backup, 1944af31f5e5SChris Mason btrfs_header_generation(info->csum_root->node)); 1945af31f5e5SChris Mason btrfs_set_backup_csum_root_level(root_backup, 1946af31f5e5SChris Mason btrfs_header_level(info->csum_root->node)); 1947af31f5e5SChris Mason 1948af31f5e5SChris Mason btrfs_set_backup_total_bytes(root_backup, 1949af31f5e5SChris Mason btrfs_super_total_bytes(info->super_copy)); 1950af31f5e5SChris Mason btrfs_set_backup_bytes_used(root_backup, 1951af31f5e5SChris Mason btrfs_super_bytes_used(info->super_copy)); 1952af31f5e5SChris Mason btrfs_set_backup_num_devices(root_backup, 1953af31f5e5SChris Mason btrfs_super_num_devices(info->super_copy)); 1954af31f5e5SChris Mason 1955af31f5e5SChris Mason /* 1956af31f5e5SChris Mason * if we don't copy this out to the super_copy, it won't get remembered 1957af31f5e5SChris Mason * for the next commit 1958af31f5e5SChris Mason */ 1959af31f5e5SChris Mason memcpy(&info->super_copy->super_roots, 1960af31f5e5SChris Mason &info->super_for_commit->super_roots, 1961af31f5e5SChris Mason sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1962af31f5e5SChris Mason } 1963af31f5e5SChris Mason 1964af31f5e5SChris Mason /* 1965af31f5e5SChris Mason * this copies info out of the root backup array and back into 1966af31f5e5SChris Mason * the in-memory super block. It is meant to help iterate through 1967af31f5e5SChris Mason * the array, so you send it the number of backups you've already 1968af31f5e5SChris Mason * tried and the last backup index you used. 1969af31f5e5SChris Mason * 1970af31f5e5SChris Mason * this returns -1 when it has tried all the backups 1971af31f5e5SChris Mason */ 1972af31f5e5SChris Mason static noinline int next_root_backup(struct btrfs_fs_info *info, 1973af31f5e5SChris Mason struct btrfs_super_block *super, 1974af31f5e5SChris Mason int *num_backups_tried, int *backup_index) 1975af31f5e5SChris Mason { 1976af31f5e5SChris Mason struct btrfs_root_backup *root_backup; 1977af31f5e5SChris Mason int newest = *backup_index; 1978af31f5e5SChris Mason 1979af31f5e5SChris Mason if (*num_backups_tried == 0) { 1980af31f5e5SChris Mason u64 gen = btrfs_super_generation(super); 1981af31f5e5SChris Mason 1982af31f5e5SChris Mason newest = find_newest_super_backup(info, gen); 1983af31f5e5SChris Mason if (newest == -1) 1984af31f5e5SChris Mason return -1; 1985af31f5e5SChris Mason 1986af31f5e5SChris Mason *backup_index = newest; 1987af31f5e5SChris Mason *num_backups_tried = 1; 1988af31f5e5SChris Mason } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 1989af31f5e5SChris Mason /* we've tried all the backups, all done */ 1990af31f5e5SChris Mason return -1; 1991af31f5e5SChris Mason } else { 1992af31f5e5SChris Mason /* jump to the next oldest backup */ 1993af31f5e5SChris Mason newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 1994af31f5e5SChris Mason BTRFS_NUM_BACKUP_ROOTS; 1995af31f5e5SChris Mason *backup_index = newest; 1996af31f5e5SChris Mason *num_backups_tried += 1; 1997af31f5e5SChris Mason } 1998af31f5e5SChris Mason root_backup = super->super_roots + newest; 1999af31f5e5SChris Mason 2000af31f5e5SChris Mason btrfs_set_super_generation(super, 2001af31f5e5SChris Mason btrfs_backup_tree_root_gen(root_backup)); 2002af31f5e5SChris Mason btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2003af31f5e5SChris Mason btrfs_set_super_root_level(super, 2004af31f5e5SChris Mason btrfs_backup_tree_root_level(root_backup)); 2005af31f5e5SChris Mason btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2006af31f5e5SChris Mason 2007af31f5e5SChris Mason /* 2008af31f5e5SChris Mason * fixme: the total bytes and num_devices need to match or we should 2009af31f5e5SChris Mason * need a fsck 2010af31f5e5SChris Mason */ 2011af31f5e5SChris Mason btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2012af31f5e5SChris Mason btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2013af31f5e5SChris Mason return 0; 2014af31f5e5SChris Mason } 2015af31f5e5SChris Mason 20167abadb64SLiu Bo /* helper to cleanup workers */ 20177abadb64SLiu Bo static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 20187abadb64SLiu Bo { 20197abadb64SLiu Bo btrfs_stop_workers(&fs_info->generic_worker); 20207abadb64SLiu Bo btrfs_stop_workers(&fs_info->fixup_workers); 20217abadb64SLiu Bo btrfs_stop_workers(&fs_info->delalloc_workers); 20227abadb64SLiu Bo btrfs_stop_workers(&fs_info->workers); 20237abadb64SLiu Bo btrfs_stop_workers(&fs_info->endio_workers); 20247abadb64SLiu Bo btrfs_stop_workers(&fs_info->endio_meta_workers); 20257abadb64SLiu Bo btrfs_stop_workers(&fs_info->endio_raid56_workers); 20267abadb64SLiu Bo btrfs_stop_workers(&fs_info->rmw_workers); 20277abadb64SLiu Bo btrfs_stop_workers(&fs_info->endio_meta_write_workers); 20287abadb64SLiu Bo btrfs_stop_workers(&fs_info->endio_write_workers); 20297abadb64SLiu Bo btrfs_stop_workers(&fs_info->endio_freespace_worker); 20307abadb64SLiu Bo btrfs_stop_workers(&fs_info->submit_workers); 20317abadb64SLiu Bo btrfs_stop_workers(&fs_info->delayed_workers); 20327abadb64SLiu Bo btrfs_stop_workers(&fs_info->caching_workers); 20337abadb64SLiu Bo btrfs_stop_workers(&fs_info->readahead_workers); 20347abadb64SLiu Bo btrfs_stop_workers(&fs_info->flush_workers); 20352f232036SJan Schmidt btrfs_stop_workers(&fs_info->qgroup_rescan_workers); 20367abadb64SLiu Bo } 20377abadb64SLiu Bo 2038af31f5e5SChris Mason /* helper to cleanup tree roots */ 2039af31f5e5SChris Mason static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) 2040af31f5e5SChris Mason { 2041af31f5e5SChris Mason free_extent_buffer(info->tree_root->node); 2042af31f5e5SChris Mason free_extent_buffer(info->tree_root->commit_root); 2043655b09feSJosef Bacik info->tree_root->node = NULL; 2044655b09feSJosef Bacik info->tree_root->commit_root = NULL; 2045655b09feSJosef Bacik 2046655b09feSJosef Bacik if (info->dev_root) { 2047af31f5e5SChris Mason free_extent_buffer(info->dev_root->node); 2048af31f5e5SChris Mason free_extent_buffer(info->dev_root->commit_root); 2049655b09feSJosef Bacik info->dev_root->node = NULL; 2050655b09feSJosef Bacik info->dev_root->commit_root = NULL; 2051655b09feSJosef Bacik } 2052655b09feSJosef Bacik if (info->extent_root) { 2053af31f5e5SChris Mason free_extent_buffer(info->extent_root->node); 2054af31f5e5SChris Mason free_extent_buffer(info->extent_root->commit_root); 2055655b09feSJosef Bacik info->extent_root->node = NULL; 2056655b09feSJosef Bacik info->extent_root->commit_root = NULL; 2057655b09feSJosef Bacik } 2058655b09feSJosef Bacik if (info->csum_root) { 2059af31f5e5SChris Mason free_extent_buffer(info->csum_root->node); 2060af31f5e5SChris Mason free_extent_buffer(info->csum_root->commit_root); 2061655b09feSJosef Bacik info->csum_root->node = NULL; 2062655b09feSJosef Bacik info->csum_root->commit_root = NULL; 2063655b09feSJosef Bacik } 2064bcef60f2SArne Jansen if (info->quota_root) { 2065bcef60f2SArne Jansen free_extent_buffer(info->quota_root->node); 2066bcef60f2SArne Jansen free_extent_buffer(info->quota_root->commit_root); 2067bcef60f2SArne Jansen info->quota_root->node = NULL; 2068bcef60f2SArne Jansen info->quota_root->commit_root = NULL; 2069bcef60f2SArne Jansen } 2070f7a81ea4SStefan Behrens if (info->uuid_root) { 2071f7a81ea4SStefan Behrens free_extent_buffer(info->uuid_root->node); 2072f7a81ea4SStefan Behrens free_extent_buffer(info->uuid_root->commit_root); 2073f7a81ea4SStefan Behrens info->uuid_root->node = NULL; 2074f7a81ea4SStefan Behrens info->uuid_root->commit_root = NULL; 2075f7a81ea4SStefan Behrens } 2076af31f5e5SChris Mason if (chunk_root) { 2077af31f5e5SChris Mason free_extent_buffer(info->chunk_root->node); 2078af31f5e5SChris Mason free_extent_buffer(info->chunk_root->commit_root); 2079af31f5e5SChris Mason info->chunk_root->node = NULL; 2080af31f5e5SChris Mason info->chunk_root->commit_root = NULL; 2081af31f5e5SChris Mason } 2082af31f5e5SChris Mason } 2083af31f5e5SChris Mason 2084171f6537SJosef Bacik static void del_fs_roots(struct btrfs_fs_info *fs_info) 2085171f6537SJosef Bacik { 2086171f6537SJosef Bacik int ret; 2087171f6537SJosef Bacik struct btrfs_root *gang[8]; 2088171f6537SJosef Bacik int i; 2089171f6537SJosef Bacik 2090171f6537SJosef Bacik while (!list_empty(&fs_info->dead_roots)) { 2091171f6537SJosef Bacik gang[0] = list_entry(fs_info->dead_roots.next, 2092171f6537SJosef Bacik struct btrfs_root, root_list); 2093171f6537SJosef Bacik list_del(&gang[0]->root_list); 2094171f6537SJosef Bacik 2095171f6537SJosef Bacik if (gang[0]->in_radix) { 2096cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2097171f6537SJosef Bacik } else { 2098171f6537SJosef Bacik free_extent_buffer(gang[0]->node); 2099171f6537SJosef Bacik free_extent_buffer(gang[0]->commit_root); 2100b0feb9d9SMiao Xie btrfs_put_fs_root(gang[0]); 2101171f6537SJosef Bacik } 2102171f6537SJosef Bacik } 2103171f6537SJosef Bacik 2104171f6537SJosef Bacik while (1) { 2105171f6537SJosef Bacik ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2106171f6537SJosef Bacik (void **)gang, 0, 2107171f6537SJosef Bacik ARRAY_SIZE(gang)); 2108171f6537SJosef Bacik if (!ret) 2109171f6537SJosef Bacik break; 2110171f6537SJosef Bacik for (i = 0; i < ret; i++) 2111cb517eabSMiao Xie btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2112171f6537SJosef Bacik } 2113171f6537SJosef Bacik } 2114af31f5e5SChris Mason 2115ad2b2c80SAl Viro int open_ctree(struct super_block *sb, 2116dfe25020SChris Mason struct btrfs_fs_devices *fs_devices, 2117dfe25020SChris Mason char *options) 2118eb60ceacSChris Mason { 2119db94535dSChris Mason u32 sectorsize; 2120db94535dSChris Mason u32 nodesize; 2121db94535dSChris Mason u32 leafsize; 2122db94535dSChris Mason u32 blocksize; 212387ee04ebSChris Mason u32 stripesize; 212484234f3aSYan Zheng u64 generation; 2125f2b636e8SJosef Bacik u64 features; 21263de4586cSChris Mason struct btrfs_key location; 2127a061fc8dSChris Mason struct buffer_head *bh; 21284d34b278SIlya Dryomov struct btrfs_super_block *disk_super; 2129815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2130f84a8bd6SAl Viro struct btrfs_root *tree_root; 21314d34b278SIlya Dryomov struct btrfs_root *extent_root; 21324d34b278SIlya Dryomov struct btrfs_root *csum_root; 21334d34b278SIlya Dryomov struct btrfs_root *chunk_root; 21344d34b278SIlya Dryomov struct btrfs_root *dev_root; 2135bcef60f2SArne Jansen struct btrfs_root *quota_root; 2136f7a81ea4SStefan Behrens struct btrfs_root *uuid_root; 2137e02119d5SChris Mason struct btrfs_root *log_tree_root; 2138eb60ceacSChris Mason int ret; 2139e58ca020SYan int err = -EINVAL; 2140af31f5e5SChris Mason int num_backups_tried = 0; 2141af31f5e5SChris Mason int backup_index = 0; 214270f80175SStefan Behrens bool create_uuid_tree; 214370f80175SStefan Behrens bool check_uuid_tree; 21444543df7eSChris Mason 2145f84a8bd6SAl Viro tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info); 21466f07e42eSAl Viro chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info); 2147cb517eabSMiao Xie if (!tree_root || !chunk_root) { 214839279cc3SChris Mason err = -ENOMEM; 214939279cc3SChris Mason goto fail; 215039279cc3SChris Mason } 215176dda93cSYan, Zheng 215276dda93cSYan, Zheng ret = init_srcu_struct(&fs_info->subvol_srcu); 215376dda93cSYan, Zheng if (ret) { 215476dda93cSYan, Zheng err = ret; 215576dda93cSYan, Zheng goto fail; 215676dda93cSYan, Zheng } 215776dda93cSYan, Zheng 215876dda93cSYan, Zheng ret = setup_bdi(fs_info, &fs_info->bdi); 215976dda93cSYan, Zheng if (ret) { 216076dda93cSYan, Zheng err = ret; 216176dda93cSYan, Zheng goto fail_srcu; 216276dda93cSYan, Zheng } 216376dda93cSYan, Zheng 2164e2d84521SMiao Xie ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0); 2165e2d84521SMiao Xie if (ret) { 2166e2d84521SMiao Xie err = ret; 2167e2d84521SMiao Xie goto fail_bdi; 2168e2d84521SMiao Xie } 2169e2d84521SMiao Xie fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE * 2170e2d84521SMiao Xie (1 + ilog2(nr_cpu_ids)); 2171e2d84521SMiao Xie 2172963d678bSMiao Xie ret = percpu_counter_init(&fs_info->delalloc_bytes, 0); 2173963d678bSMiao Xie if (ret) { 2174963d678bSMiao Xie err = ret; 2175963d678bSMiao Xie goto fail_dirty_metadata_bytes; 2176963d678bSMiao Xie } 2177963d678bSMiao Xie 217876dda93cSYan, Zheng fs_info->btree_inode = new_inode(sb); 217976dda93cSYan, Zheng if (!fs_info->btree_inode) { 218076dda93cSYan, Zheng err = -ENOMEM; 2181963d678bSMiao Xie goto fail_delalloc_bytes; 218276dda93cSYan, Zheng } 218376dda93cSYan, Zheng 2184a6591715SChris Mason mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 21851561dedaSMiao Xie 218676dda93cSYan, Zheng INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 21878fd17795SChris Mason INIT_LIST_HEAD(&fs_info->trans_list); 2188facda1e7SChris Mason INIT_LIST_HEAD(&fs_info->dead_roots); 218924bbcf04SYan, Zheng INIT_LIST_HEAD(&fs_info->delayed_iputs); 2190eb73c1b7SMiao Xie INIT_LIST_HEAD(&fs_info->delalloc_roots); 219111833d66SYan Zheng INIT_LIST_HEAD(&fs_info->caching_block_groups); 2192eb73c1b7SMiao Xie spin_lock_init(&fs_info->delalloc_root_lock); 2193a4abeea4SJosef Bacik spin_lock_init(&fs_info->trans_lock); 219476dda93cSYan, Zheng spin_lock_init(&fs_info->fs_roots_radix_lock); 219524bbcf04SYan, Zheng spin_lock_init(&fs_info->delayed_iput_lock); 21964cb5300bSChris Mason spin_lock_init(&fs_info->defrag_inodes_lock); 21972bf64758SJosef Bacik spin_lock_init(&fs_info->free_chunk_lock); 2198f29021b2SJan Schmidt spin_lock_init(&fs_info->tree_mod_seq_lock); 2199ceda0864SMiao Xie spin_lock_init(&fs_info->super_lock); 2200f29021b2SJan Schmidt rwlock_init(&fs_info->tree_mod_log_lock); 22017585717fSChris Mason mutex_init(&fs_info->reloc_mutex); 2202de98ced9SMiao Xie seqlock_init(&fs_info->profiles_lock); 220319c00ddcSChris Mason 220458176a96SJosef Bacik init_completion(&fs_info->kobj_unregister); 22050b86a832SChris Mason INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 22066324fbf3SChris Mason INIT_LIST_HEAD(&fs_info->space_info); 2207f29021b2SJan Schmidt INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 22080b86a832SChris Mason btrfs_mapping_init(&fs_info->mapping_tree); 220966d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->global_block_rsv, 221066d8f3ddSMiao Xie BTRFS_BLOCK_RSV_GLOBAL); 221166d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delalloc_block_rsv, 221266d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELALLOC); 221366d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 221466d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 221566d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 221666d8f3ddSMiao Xie btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 221766d8f3ddSMiao Xie BTRFS_BLOCK_RSV_DELOPS); 2218cb03c743SChris Mason atomic_set(&fs_info->nr_async_submits, 0); 2219771ed689SChris Mason atomic_set(&fs_info->async_delalloc_pages, 0); 22208c8bee1dSChris Mason atomic_set(&fs_info->async_submit_draining, 0); 22210986fe9eSChris Mason atomic_set(&fs_info->nr_async_bios, 0); 22224cb5300bSChris Mason atomic_set(&fs_info->defrag_running, 0); 2223fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 2224e20d96d6SChris Mason fs_info->sb = sb; 22256f568d35SChris Mason fs_info->max_inline = 8192 * 1024; 22269ed74f2dSJosef Bacik fs_info->metadata_ratio = 0; 22274cb5300bSChris Mason fs_info->defrag_inodes = RB_ROOT; 22282bf64758SJosef Bacik fs_info->free_chunk_space = 0; 2229f29021b2SJan Schmidt fs_info->tree_mod_log = RB_ROOT; 22308b87dc17SDavid Sterba fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2231c8b97818SChris Mason 223290519d66SArne Jansen /* readahead state */ 223390519d66SArne Jansen INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT); 223490519d66SArne Jansen spin_lock_init(&fs_info->reada_lock); 22352c90e5d6SChris Mason 2236b34b086cSChris Mason fs_info->thread_pool_size = min_t(unsigned long, 2237b34b086cSChris Mason num_online_cpus() + 2, 8); 22380afbaf8cSChris Mason 2239199c2a9cSMiao Xie INIT_LIST_HEAD(&fs_info->ordered_roots); 2240199c2a9cSMiao Xie spin_lock_init(&fs_info->ordered_root_lock); 224116cdcec7SMiao Xie fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 224216cdcec7SMiao Xie GFP_NOFS); 224316cdcec7SMiao Xie if (!fs_info->delayed_root) { 224416cdcec7SMiao Xie err = -ENOMEM; 224516cdcec7SMiao Xie goto fail_iput; 224616cdcec7SMiao Xie } 224716cdcec7SMiao Xie btrfs_init_delayed_root(fs_info->delayed_root); 22483eaa2885SChris Mason 2249a2de733cSArne Jansen mutex_init(&fs_info->scrub_lock); 2250a2de733cSArne Jansen atomic_set(&fs_info->scrubs_running, 0); 2251a2de733cSArne Jansen atomic_set(&fs_info->scrub_pause_req, 0); 2252a2de733cSArne Jansen atomic_set(&fs_info->scrubs_paused, 0); 2253a2de733cSArne Jansen atomic_set(&fs_info->scrub_cancel_req, 0); 2254a2de733cSArne Jansen init_waitqueue_head(&fs_info->scrub_pause_wait); 2255a2de733cSArne Jansen init_rwsem(&fs_info->scrub_super_lock); 2256a2de733cSArne Jansen fs_info->scrub_workers_refcnt = 0; 225721adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 225821adbd5cSStefan Behrens fs_info->check_integrity_print_mask = 0; 225921adbd5cSStefan Behrens #endif 2260a2de733cSArne Jansen 2261c9e9f97bSIlya Dryomov spin_lock_init(&fs_info->balance_lock); 2262c9e9f97bSIlya Dryomov mutex_init(&fs_info->balance_mutex); 2263837d5b6eSIlya Dryomov atomic_set(&fs_info->balance_running, 0); 2264837d5b6eSIlya Dryomov atomic_set(&fs_info->balance_pause_req, 0); 2265a7e99c69SIlya Dryomov atomic_set(&fs_info->balance_cancel_req, 0); 2266c9e9f97bSIlya Dryomov fs_info->balance_ctl = NULL; 2267837d5b6eSIlya Dryomov init_waitqueue_head(&fs_info->balance_wait_q); 2268a061fc8dSChris Mason 22690afbaf8cSChris Mason sb->s_blocksize = 4096; 22700afbaf8cSChris Mason sb->s_blocksize_bits = blksize_bits(4096); 227132a88aa1SJens Axboe sb->s_bdi = &fs_info->bdi; 22720afbaf8cSChris Mason 227376dda93cSYan, Zheng fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 2274bfe86848SMiklos Szeredi set_nlink(fs_info->btree_inode, 1); 22750afbaf8cSChris Mason /* 22760afbaf8cSChris Mason * we set the i_size on the btree inode to the max possible int. 22770afbaf8cSChris Mason * the real end of the address space is determined by all of 22780afbaf8cSChris Mason * the devices in the system 22790afbaf8cSChris Mason */ 22800afbaf8cSChris Mason fs_info->btree_inode->i_size = OFFSET_MAX; 2281d98237b3SChris Mason fs_info->btree_inode->i_mapping->a_ops = &btree_aops; 228204160088SChris Mason fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi; 228304160088SChris Mason 22845d4f98a2SYan Zheng RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node); 2285d1310b2eSChris Mason extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree, 2286f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 22870b32f4bbSJosef Bacik BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0; 2288a8067e02SDavid Sterba extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree); 22890da5468fSChris Mason 2290d1310b2eSChris Mason BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 2291d1310b2eSChris Mason 22920f7d52f4SChris Mason BTRFS_I(fs_info->btree_inode)->root = tree_root; 22930f7d52f4SChris Mason memset(&BTRFS_I(fs_info->btree_inode)->location, 0, 22940f7d52f4SChris Mason sizeof(struct btrfs_key)); 229572ac3c0dSJosef Bacik set_bit(BTRFS_INODE_DUMMY, 229672ac3c0dSJosef Bacik &BTRFS_I(fs_info->btree_inode)->runtime_flags); 2297778ba82bSFilipe David Borba Manana btrfs_insert_inode_hash(fs_info->btree_inode); 229839279cc3SChris Mason 2299e02119d5SChris Mason spin_lock_init(&fs_info->block_group_cache_lock); 23006bef4d31SEric Paris fs_info->block_group_cache_tree = RB_ROOT; 2301a1897fddSLiu Bo fs_info->first_logical_byte = (u64)-1; 2302925baeddSChris Mason 230311833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[0], 2304f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 230511833d66SYan Zheng extent_io_tree_init(&fs_info->freed_extents[1], 2306f993c883SDavid Sterba fs_info->btree_inode->i_mapping); 230711833d66SYan Zheng fs_info->pinned_extents = &fs_info->freed_extents[0]; 2308e6dcd2dcSChris Mason fs_info->do_barriers = 1; 2309f9295749SChris Mason 2310d98237b3SChris Mason 23115a3f23d5SChris Mason mutex_init(&fs_info->ordered_operations_mutex); 23129ffba8cdSJosef Bacik mutex_init(&fs_info->ordered_extent_flush_mutex); 231330ae8467SChris Mason mutex_init(&fs_info->tree_log_mutex); 2314925baeddSChris Mason mutex_init(&fs_info->chunk_mutex); 2315925baeddSChris Mason mutex_init(&fs_info->transaction_kthread_mutex); 2316a74a4b97SChris Mason mutex_init(&fs_info->cleaner_mutex); 2317a74a4b97SChris Mason mutex_init(&fs_info->volume_mutex); 2318276e680dSYan Zheng init_rwsem(&fs_info->extent_commit_sem); 2319c71bf099SYan, Zheng init_rwsem(&fs_info->cleanup_work_sem); 232076dda93cSYan, Zheng init_rwsem(&fs_info->subvol_sem); 2321803b2f54SStefan Behrens sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2322e922e087SStefan Behrens fs_info->dev_replace.lock_owner = 0; 2323e922e087SStefan Behrens atomic_set(&fs_info->dev_replace.nesting_level, 0); 2324e922e087SStefan Behrens mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2325e922e087SStefan Behrens mutex_init(&fs_info->dev_replace.lock_management_lock); 2326e922e087SStefan Behrens mutex_init(&fs_info->dev_replace.lock); 2327fa9c0d79SChris Mason 2328416ac51dSArne Jansen spin_lock_init(&fs_info->qgroup_lock); 2329f2f6ed3dSWang Shilong mutex_init(&fs_info->qgroup_ioctl_lock); 2330416ac51dSArne Jansen fs_info->qgroup_tree = RB_ROOT; 2331416ac51dSArne Jansen INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2332416ac51dSArne Jansen fs_info->qgroup_seq = 1; 2333416ac51dSArne Jansen fs_info->quota_enabled = 0; 2334416ac51dSArne Jansen fs_info->pending_quota_state = 0; 23351e8f9158SWang Shilong fs_info->qgroup_ulist = NULL; 23362f232036SJan Schmidt mutex_init(&fs_info->qgroup_rescan_lock); 2337416ac51dSArne Jansen 2338fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2339fa9c0d79SChris Mason btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2340fa9c0d79SChris Mason 23417d9eb12cSChris Mason init_waitqueue_head(&fs_info->transaction_throttle); 23423768f368SChris Mason init_waitqueue_head(&fs_info->transaction_wait); 2343bb9c12c9SSage Weil init_waitqueue_head(&fs_info->transaction_blocked_wait); 23444854ddd0SChris Mason init_waitqueue_head(&fs_info->async_submit_wait); 23459a8dd150SChris Mason 234653b381b3SDavid Woodhouse ret = btrfs_alloc_stripe_hash_table(fs_info); 234753b381b3SDavid Woodhouse if (ret) { 234883c8266aSDavid Sterba err = ret; 234953b381b3SDavid Woodhouse goto fail_alloc; 235053b381b3SDavid Woodhouse } 235153b381b3SDavid Woodhouse 23520b86a832SChris Mason __setup_root(4096, 4096, 4096, 4096, tree_root, 23532c90e5d6SChris Mason fs_info, BTRFS_ROOT_TREE_OBJECTID); 23547eccb903SChris Mason 23553c4bb26bSChris Mason invalidate_bdev(fs_devices->latest_bdev); 23561104a885SDavid Sterba 23571104a885SDavid Sterba /* 23581104a885SDavid Sterba * Read super block and check the signature bytes only 23591104a885SDavid Sterba */ 2360a512bbf8SYan Zheng bh = btrfs_read_dev_super(fs_devices->latest_bdev); 236120b45077SDave Young if (!bh) { 236220b45077SDave Young err = -EINVAL; 236316cdcec7SMiao Xie goto fail_alloc; 236420b45077SDave Young } 236539279cc3SChris Mason 23661104a885SDavid Sterba /* 23671104a885SDavid Sterba * We want to check superblock checksum, the type is stored inside. 23681104a885SDavid Sterba * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 23691104a885SDavid Sterba */ 23701104a885SDavid Sterba if (btrfs_check_super_csum(bh->b_data)) { 23711104a885SDavid Sterba printk(KERN_ERR "btrfs: superblock checksum mismatch\n"); 23721104a885SDavid Sterba err = -EINVAL; 23731104a885SDavid Sterba goto fail_alloc; 23741104a885SDavid Sterba } 23751104a885SDavid Sterba 23761104a885SDavid Sterba /* 23771104a885SDavid Sterba * super_copy is zeroed at allocation time and we never touch the 23781104a885SDavid Sterba * following bytes up to INFO_SIZE, the checksum is calculated from 23791104a885SDavid Sterba * the whole block of INFO_SIZE 23801104a885SDavid Sterba */ 23816c41761fSDavid Sterba memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 23826c41761fSDavid Sterba memcpy(fs_info->super_for_commit, fs_info->super_copy, 23836c41761fSDavid Sterba sizeof(*fs_info->super_for_commit)); 2384a061fc8dSChris Mason brelse(bh); 23855f39d397SChris Mason 23866c41761fSDavid Sterba memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE); 23870b86a832SChris Mason 23881104a885SDavid Sterba ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY); 23891104a885SDavid Sterba if (ret) { 23901104a885SDavid Sterba printk(KERN_ERR "btrfs: superblock contains fatal errors\n"); 23911104a885SDavid Sterba err = -EINVAL; 23921104a885SDavid Sterba goto fail_alloc; 23931104a885SDavid Sterba } 23941104a885SDavid Sterba 23956c41761fSDavid Sterba disk_super = fs_info->super_copy; 23960f7d52f4SChris Mason if (!btrfs_super_root(disk_super)) 239716cdcec7SMiao Xie goto fail_alloc; 23980f7d52f4SChris Mason 2399acce952bSliubo /* check FS state, whether FS is broken. */ 240087533c47SMiao Xie if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 240187533c47SMiao Xie set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2402acce952bSliubo 240375e7cb7fSLiu Bo /* 2404af31f5e5SChris Mason * run through our array of backup supers and setup 2405af31f5e5SChris Mason * our ring pointer to the oldest one 2406af31f5e5SChris Mason */ 2407af31f5e5SChris Mason generation = btrfs_super_generation(disk_super); 2408af31f5e5SChris Mason find_oldest_super_backup(fs_info, generation); 2409af31f5e5SChris Mason 2410af31f5e5SChris Mason /* 241175e7cb7fSLiu Bo * In the long term, we'll store the compression type in the super 241275e7cb7fSLiu Bo * block, and it'll be used for per file compression control. 241375e7cb7fSLiu Bo */ 241475e7cb7fSLiu Bo fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 241575e7cb7fSLiu Bo 24162b82032cSYan Zheng ret = btrfs_parse_options(tree_root, options); 24172b82032cSYan Zheng if (ret) { 24182b82032cSYan Zheng err = ret; 241916cdcec7SMiao Xie goto fail_alloc; 24202b82032cSYan Zheng } 2421dfe25020SChris Mason 2422f2b636e8SJosef Bacik features = btrfs_super_incompat_flags(disk_super) & 2423f2b636e8SJosef Bacik ~BTRFS_FEATURE_INCOMPAT_SUPP; 2424f2b636e8SJosef Bacik if (features) { 2425f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount because of " 2426f2b636e8SJosef Bacik "unsupported optional features (%Lx).\n", 2427c1c9ff7cSGeert Uytterhoeven features); 2428f2b636e8SJosef Bacik err = -EINVAL; 242916cdcec7SMiao Xie goto fail_alloc; 2430f2b636e8SJosef Bacik } 2431f2b636e8SJosef Bacik 2432727011e0SChris Mason if (btrfs_super_leafsize(disk_super) != 2433727011e0SChris Mason btrfs_super_nodesize(disk_super)) { 2434727011e0SChris Mason printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2435727011e0SChris Mason "blocksizes don't match. node %d leaf %d\n", 2436727011e0SChris Mason btrfs_super_nodesize(disk_super), 2437727011e0SChris Mason btrfs_super_leafsize(disk_super)); 2438727011e0SChris Mason err = -EINVAL; 2439727011e0SChris Mason goto fail_alloc; 2440727011e0SChris Mason } 2441727011e0SChris Mason if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) { 2442727011e0SChris Mason printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2443727011e0SChris Mason "blocksize (%d) was too large\n", 2444727011e0SChris Mason btrfs_super_leafsize(disk_super)); 2445727011e0SChris Mason err = -EINVAL; 2446727011e0SChris Mason goto fail_alloc; 2447727011e0SChris Mason } 2448727011e0SChris Mason 24495d4f98a2SYan Zheng features = btrfs_super_incompat_flags(disk_super); 24505d4f98a2SYan Zheng features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 245169e380d1SLi Zefan if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO) 2452a6fa6faeSLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 2453727011e0SChris Mason 24543173a18fSJosef Bacik if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 24553173a18fSJosef Bacik printk(KERN_ERR "btrfs: has skinny extents\n"); 24563173a18fSJosef Bacik 2457727011e0SChris Mason /* 2458727011e0SChris Mason * flag our filesystem as having big metadata blocks if 2459727011e0SChris Mason * they are bigger than the page size 2460727011e0SChris Mason */ 2461727011e0SChris Mason if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) { 2462727011e0SChris Mason if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 2463727011e0SChris Mason printk(KERN_INFO "btrfs flagging fs with big metadata feature\n"); 2464727011e0SChris Mason features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2465727011e0SChris Mason } 2466727011e0SChris Mason 2467bc3f116fSChris Mason nodesize = btrfs_super_nodesize(disk_super); 2468bc3f116fSChris Mason leafsize = btrfs_super_leafsize(disk_super); 2469bc3f116fSChris Mason sectorsize = btrfs_super_sectorsize(disk_super); 2470bc3f116fSChris Mason stripesize = btrfs_super_stripesize(disk_super); 2471e2d84521SMiao Xie fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids)); 2472963d678bSMiao Xie fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2473bc3f116fSChris Mason 2474bc3f116fSChris Mason /* 2475bc3f116fSChris Mason * mixed block groups end up with duplicate but slightly offset 2476bc3f116fSChris Mason * extent buffers for the same range. It leads to corruptions 2477bc3f116fSChris Mason */ 2478bc3f116fSChris Mason if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2479bc3f116fSChris Mason (sectorsize != leafsize)) { 2480bc3f116fSChris Mason printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes " 2481bc3f116fSChris Mason "are not allowed for mixed block groups on %s\n", 2482bc3f116fSChris Mason sb->s_id); 2483bc3f116fSChris Mason goto fail_alloc; 2484bc3f116fSChris Mason } 2485bc3f116fSChris Mason 2486ceda0864SMiao Xie /* 2487ceda0864SMiao Xie * Needn't use the lock because there is no other task which will 2488ceda0864SMiao Xie * update the flag. 2489ceda0864SMiao Xie */ 24905d4f98a2SYan Zheng btrfs_set_super_incompat_flags(disk_super, features); 24915d4f98a2SYan Zheng 2492f2b636e8SJosef Bacik features = btrfs_super_compat_ro_flags(disk_super) & 2493f2b636e8SJosef Bacik ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2494f2b636e8SJosef Bacik if (!(sb->s_flags & MS_RDONLY) && features) { 2495f2b636e8SJosef Bacik printk(KERN_ERR "BTRFS: couldn't mount RDWR because of " 2496f2b636e8SJosef Bacik "unsupported option features (%Lx).\n", 2497c1c9ff7cSGeert Uytterhoeven features); 2498f2b636e8SJosef Bacik err = -EINVAL; 249916cdcec7SMiao Xie goto fail_alloc; 2500f2b636e8SJosef Bacik } 250161d92c32SChris Mason 250261d92c32SChris Mason btrfs_init_workers(&fs_info->generic_worker, 250361d92c32SChris Mason "genwork", 1, NULL); 250461d92c32SChris Mason 25055443be45SChris Mason btrfs_init_workers(&fs_info->workers, "worker", 250661d92c32SChris Mason fs_info->thread_pool_size, 250761d92c32SChris Mason &fs_info->generic_worker); 2508c8b97818SChris Mason 2509771ed689SChris Mason btrfs_init_workers(&fs_info->delalloc_workers, "delalloc", 251045d5fd14SJosef Bacik fs_info->thread_pool_size, NULL); 2511771ed689SChris Mason 25128ccf6f19SMiao Xie btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc", 251345d5fd14SJosef Bacik fs_info->thread_pool_size, NULL); 25148ccf6f19SMiao Xie 25155443be45SChris Mason btrfs_init_workers(&fs_info->submit_workers, "submit", 2516b720d209SChris Mason min_t(u64, fs_devices->num_devices, 251745d5fd14SJosef Bacik fs_info->thread_pool_size), NULL); 251861b49440SChris Mason 2519bab39bf9SJosef Bacik btrfs_init_workers(&fs_info->caching_workers, "cache", 252045d5fd14SJosef Bacik fs_info->thread_pool_size, NULL); 2521bab39bf9SJosef Bacik 252261b49440SChris Mason /* a higher idle thresh on the submit workers makes it much more 252361b49440SChris Mason * likely that bios will be send down in a sane order to the 252461b49440SChris Mason * devices 252561b49440SChris Mason */ 252661b49440SChris Mason fs_info->submit_workers.idle_thresh = 64; 252753863232SChris Mason 2528771ed689SChris Mason fs_info->workers.idle_thresh = 16; 25294a69a410SChris Mason fs_info->workers.ordered = 1; 253061b49440SChris Mason 2531771ed689SChris Mason fs_info->delalloc_workers.idle_thresh = 2; 2532771ed689SChris Mason fs_info->delalloc_workers.ordered = 1; 2533771ed689SChris Mason 253461d92c32SChris Mason btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1, 253561d92c32SChris Mason &fs_info->generic_worker); 25365443be45SChris Mason btrfs_init_workers(&fs_info->endio_workers, "endio", 253761d92c32SChris Mason fs_info->thread_pool_size, 253861d92c32SChris Mason &fs_info->generic_worker); 2539d20f7043SChris Mason btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta", 254061d92c32SChris Mason fs_info->thread_pool_size, 254161d92c32SChris Mason &fs_info->generic_worker); 2542cad321adSChris Mason btrfs_init_workers(&fs_info->endio_meta_write_workers, 254361d92c32SChris Mason "endio-meta-write", fs_info->thread_pool_size, 254461d92c32SChris Mason &fs_info->generic_worker); 254553b381b3SDavid Woodhouse btrfs_init_workers(&fs_info->endio_raid56_workers, 254653b381b3SDavid Woodhouse "endio-raid56", fs_info->thread_pool_size, 254753b381b3SDavid Woodhouse &fs_info->generic_worker); 254853b381b3SDavid Woodhouse btrfs_init_workers(&fs_info->rmw_workers, 254953b381b3SDavid Woodhouse "rmw", fs_info->thread_pool_size, 255053b381b3SDavid Woodhouse &fs_info->generic_worker); 25515443be45SChris Mason btrfs_init_workers(&fs_info->endio_write_workers, "endio-write", 255261d92c32SChris Mason fs_info->thread_pool_size, 255361d92c32SChris Mason &fs_info->generic_worker); 25540cb59c99SJosef Bacik btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write", 25550cb59c99SJosef Bacik 1, &fs_info->generic_worker); 255616cdcec7SMiao Xie btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta", 255716cdcec7SMiao Xie fs_info->thread_pool_size, 255816cdcec7SMiao Xie &fs_info->generic_worker); 255990519d66SArne Jansen btrfs_init_workers(&fs_info->readahead_workers, "readahead", 256090519d66SArne Jansen fs_info->thread_pool_size, 256190519d66SArne Jansen &fs_info->generic_worker); 25622f232036SJan Schmidt btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1, 25632f232036SJan Schmidt &fs_info->generic_worker); 256461b49440SChris Mason 256561b49440SChris Mason /* 256661b49440SChris Mason * endios are largely parallel and should have a very 256761b49440SChris Mason * low idle thresh 256861b49440SChris Mason */ 256961b49440SChris Mason fs_info->endio_workers.idle_thresh = 4; 2570b51912c9SChris Mason fs_info->endio_meta_workers.idle_thresh = 4; 257153b381b3SDavid Woodhouse fs_info->endio_raid56_workers.idle_thresh = 4; 257253b381b3SDavid Woodhouse fs_info->rmw_workers.idle_thresh = 2; 2573b51912c9SChris Mason 25749042846bSChris Mason fs_info->endio_write_workers.idle_thresh = 2; 25759042846bSChris Mason fs_info->endio_meta_write_workers.idle_thresh = 2; 257690519d66SArne Jansen fs_info->readahead_workers.idle_thresh = 2; 25779042846bSChris Mason 25780dc3b84aSJosef Bacik /* 25790dc3b84aSJosef Bacik * btrfs_start_workers can really only fail because of ENOMEM so just 25800dc3b84aSJosef Bacik * return -ENOMEM if any of these fail. 25810dc3b84aSJosef Bacik */ 25820dc3b84aSJosef Bacik ret = btrfs_start_workers(&fs_info->workers); 25830dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->generic_worker); 25840dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->submit_workers); 25850dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->delalloc_workers); 25860dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->fixup_workers); 25870dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_workers); 25880dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_meta_workers); 258953b381b3SDavid Woodhouse ret |= btrfs_start_workers(&fs_info->rmw_workers); 259053b381b3SDavid Woodhouse ret |= btrfs_start_workers(&fs_info->endio_raid56_workers); 25910dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers); 25920dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_write_workers); 25930dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->endio_freespace_worker); 25940dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->delayed_workers); 25950dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->caching_workers); 25960dc3b84aSJosef Bacik ret |= btrfs_start_workers(&fs_info->readahead_workers); 25978ccf6f19SMiao Xie ret |= btrfs_start_workers(&fs_info->flush_workers); 25982f232036SJan Schmidt ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers); 25990dc3b84aSJosef Bacik if (ret) { 2600fed425c7SIlya Dryomov err = -ENOMEM; 26010dc3b84aSJosef Bacik goto fail_sb_buffer; 26020dc3b84aSJosef Bacik } 26034543df7eSChris Mason 26044575c9ccSChris Mason fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 2605c8b97818SChris Mason fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 2606c8b97818SChris Mason 4 * 1024 * 1024 / PAGE_CACHE_SIZE); 26074575c9ccSChris Mason 2608db94535dSChris Mason tree_root->nodesize = nodesize; 2609db94535dSChris Mason tree_root->leafsize = leafsize; 2610db94535dSChris Mason tree_root->sectorsize = sectorsize; 261187ee04ebSChris Mason tree_root->stripesize = stripesize; 2612a061fc8dSChris Mason 2613a061fc8dSChris Mason sb->s_blocksize = sectorsize; 2614a061fc8dSChris Mason sb->s_blocksize_bits = blksize_bits(sectorsize); 2615db94535dSChris Mason 26163cae210fSQu Wenruo if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 2617d397712bSChris Mason printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id); 261839279cc3SChris Mason goto fail_sb_buffer; 261939279cc3SChris Mason } 262019c00ddcSChris Mason 26218d082fb7SLiu Bo if (sectorsize != PAGE_SIZE) { 26228d082fb7SLiu Bo printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) " 26238d082fb7SLiu Bo "found on %s\n", (unsigned long)sectorsize, sb->s_id); 2624941b2ddfSKeith Mannthey goto fail_sb_buffer; 2625941b2ddfSKeith Mannthey } 2626941b2ddfSKeith Mannthey 2627925baeddSChris Mason mutex_lock(&fs_info->chunk_mutex); 2628e4404d6eSYan Zheng ret = btrfs_read_sys_array(tree_root); 2629925baeddSChris Mason mutex_unlock(&fs_info->chunk_mutex); 263084eed90fSChris Mason if (ret) { 2631d397712bSChris Mason printk(KERN_WARNING "btrfs: failed to read the system " 2632d397712bSChris Mason "array on %s\n", sb->s_id); 26335d4f98a2SYan Zheng goto fail_sb_buffer; 263484eed90fSChris Mason } 26350b86a832SChris Mason 26360b86a832SChris Mason blocksize = btrfs_level_size(tree_root, 26370b86a832SChris Mason btrfs_super_chunk_root_level(disk_super)); 263884234f3aSYan Zheng generation = btrfs_super_chunk_root_generation(disk_super); 26390b86a832SChris Mason 26400b86a832SChris Mason __setup_root(nodesize, leafsize, sectorsize, stripesize, 26410b86a832SChris Mason chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 26420b86a832SChris Mason 26430b86a832SChris Mason chunk_root->node = read_tree_block(chunk_root, 26440b86a832SChris Mason btrfs_super_chunk_root(disk_super), 264584234f3aSYan Zheng blocksize, generation); 2646416bc658SJosef Bacik if (!chunk_root->node || 2647416bc658SJosef Bacik !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) { 264883121942SDavid Woodhouse printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n", 264983121942SDavid Woodhouse sb->s_id); 2650af31f5e5SChris Mason goto fail_tree_roots; 265183121942SDavid Woodhouse } 26525d4f98a2SYan Zheng btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 26535d4f98a2SYan Zheng chunk_root->commit_root = btrfs_root_node(chunk_root); 26540b86a832SChris Mason 2655e17cade2SChris Mason read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 2656b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 2657e17cade2SChris Mason 26580b86a832SChris Mason ret = btrfs_read_chunk_tree(chunk_root); 26592b82032cSYan Zheng if (ret) { 2660d397712bSChris Mason printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n", 2661d397712bSChris Mason sb->s_id); 2662af31f5e5SChris Mason goto fail_tree_roots; 26632b82032cSYan Zheng } 26640b86a832SChris Mason 26658dabb742SStefan Behrens /* 26668dabb742SStefan Behrens * keep the device that is marked to be the target device for the 26678dabb742SStefan Behrens * dev_replace procedure 26688dabb742SStefan Behrens */ 26698dabb742SStefan Behrens btrfs_close_extra_devices(fs_info, fs_devices, 0); 2670dfe25020SChris Mason 2671a6b0d5c8SChris Mason if (!fs_devices->latest_bdev) { 2672a6b0d5c8SChris Mason printk(KERN_CRIT "btrfs: failed to read devices on %s\n", 2673a6b0d5c8SChris Mason sb->s_id); 2674a6b0d5c8SChris Mason goto fail_tree_roots; 2675a6b0d5c8SChris Mason } 2676a6b0d5c8SChris Mason 2677af31f5e5SChris Mason retry_root_backup: 2678db94535dSChris Mason blocksize = btrfs_level_size(tree_root, 2679db94535dSChris Mason btrfs_super_root_level(disk_super)); 268084234f3aSYan Zheng generation = btrfs_super_generation(disk_super); 26810b86a832SChris Mason 2682e20d96d6SChris Mason tree_root->node = read_tree_block(tree_root, 2683db94535dSChris Mason btrfs_super_root(disk_super), 268484234f3aSYan Zheng blocksize, generation); 2685af31f5e5SChris Mason if (!tree_root->node || 2686af31f5e5SChris Mason !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) { 268783121942SDavid Woodhouse printk(KERN_WARNING "btrfs: failed to read tree root on %s\n", 268883121942SDavid Woodhouse sb->s_id); 2689af31f5e5SChris Mason 2690af31f5e5SChris Mason goto recovery_tree_root; 269183121942SDavid Woodhouse } 2692af31f5e5SChris Mason 26935d4f98a2SYan Zheng btrfs_set_root_node(&tree_root->root_item, tree_root->node); 26945d4f98a2SYan Zheng tree_root->commit_root = btrfs_root_node(tree_root); 269569e9c6c6SStefan Behrens btrfs_set_root_refs(&tree_root->root_item, 1); 2696db94535dSChris Mason 2697cb517eabSMiao Xie location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 2698cb517eabSMiao Xie location.type = BTRFS_ROOT_ITEM_KEY; 2699cb517eabSMiao Xie location.offset = 0; 2700cb517eabSMiao Xie 2701cb517eabSMiao Xie extent_root = btrfs_read_tree_root(tree_root, &location); 2702cb517eabSMiao Xie if (IS_ERR(extent_root)) { 2703cb517eabSMiao Xie ret = PTR_ERR(extent_root); 2704af31f5e5SChris Mason goto recovery_tree_root; 2705cb517eabSMiao Xie } 27060b86a832SChris Mason extent_root->track_dirty = 1; 2707cb517eabSMiao Xie fs_info->extent_root = extent_root; 27080b86a832SChris Mason 2709cb517eabSMiao Xie location.objectid = BTRFS_DEV_TREE_OBJECTID; 2710cb517eabSMiao Xie dev_root = btrfs_read_tree_root(tree_root, &location); 2711cb517eabSMiao Xie if (IS_ERR(dev_root)) { 2712cb517eabSMiao Xie ret = PTR_ERR(dev_root); 2713af31f5e5SChris Mason goto recovery_tree_root; 2714cb517eabSMiao Xie } 27155d4f98a2SYan Zheng dev_root->track_dirty = 1; 2716cb517eabSMiao Xie fs_info->dev_root = dev_root; 2717cb517eabSMiao Xie btrfs_init_devices_late(fs_info); 27183768f368SChris Mason 2719cb517eabSMiao Xie location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2720cb517eabSMiao Xie csum_root = btrfs_read_tree_root(tree_root, &location); 2721cb517eabSMiao Xie if (IS_ERR(csum_root)) { 2722cb517eabSMiao Xie ret = PTR_ERR(csum_root); 2723af31f5e5SChris Mason goto recovery_tree_root; 2724cb517eabSMiao Xie } 2725d20f7043SChris Mason csum_root->track_dirty = 1; 2726cb517eabSMiao Xie fs_info->csum_root = csum_root; 2727d20f7043SChris Mason 2728cb517eabSMiao Xie location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2729cb517eabSMiao Xie quota_root = btrfs_read_tree_root(tree_root, &location); 2730cb517eabSMiao Xie if (!IS_ERR(quota_root)) { 2731bcef60f2SArne Jansen quota_root->track_dirty = 1; 2732bcef60f2SArne Jansen fs_info->quota_enabled = 1; 2733bcef60f2SArne Jansen fs_info->pending_quota_state = 1; 2734cb517eabSMiao Xie fs_info->quota_root = quota_root; 2735bcef60f2SArne Jansen } 2736bcef60f2SArne Jansen 2737f7a81ea4SStefan Behrens location.objectid = BTRFS_UUID_TREE_OBJECTID; 2738f7a81ea4SStefan Behrens uuid_root = btrfs_read_tree_root(tree_root, &location); 2739f7a81ea4SStefan Behrens if (IS_ERR(uuid_root)) { 2740f7a81ea4SStefan Behrens ret = PTR_ERR(uuid_root); 2741f7a81ea4SStefan Behrens if (ret != -ENOENT) 2742f7a81ea4SStefan Behrens goto recovery_tree_root; 2743f7a81ea4SStefan Behrens create_uuid_tree = true; 274470f80175SStefan Behrens check_uuid_tree = false; 2745f7a81ea4SStefan Behrens } else { 2746f7a81ea4SStefan Behrens uuid_root->track_dirty = 1; 2747f7a81ea4SStefan Behrens fs_info->uuid_root = uuid_root; 274870f80175SStefan Behrens create_uuid_tree = false; 274970f80175SStefan Behrens check_uuid_tree = 275070f80175SStefan Behrens generation != btrfs_super_uuid_tree_generation(disk_super); 2751f7a81ea4SStefan Behrens } 2752f7a81ea4SStefan Behrens 27538929ecfaSYan, Zheng fs_info->generation = generation; 27548929ecfaSYan, Zheng fs_info->last_trans_committed = generation; 27558929ecfaSYan, Zheng 275668310a5eSIlya Dryomov ret = btrfs_recover_balance(fs_info); 275768310a5eSIlya Dryomov if (ret) { 275868310a5eSIlya Dryomov printk(KERN_WARNING "btrfs: failed to recover balance\n"); 275968310a5eSIlya Dryomov goto fail_block_groups; 276068310a5eSIlya Dryomov } 276168310a5eSIlya Dryomov 2762733f4fbbSStefan Behrens ret = btrfs_init_dev_stats(fs_info); 2763733f4fbbSStefan Behrens if (ret) { 2764733f4fbbSStefan Behrens printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n", 2765733f4fbbSStefan Behrens ret); 2766733f4fbbSStefan Behrens goto fail_block_groups; 2767733f4fbbSStefan Behrens } 2768733f4fbbSStefan Behrens 27698dabb742SStefan Behrens ret = btrfs_init_dev_replace(fs_info); 27708dabb742SStefan Behrens if (ret) { 27718dabb742SStefan Behrens pr_err("btrfs: failed to init dev_replace: %d\n", ret); 27728dabb742SStefan Behrens goto fail_block_groups; 27738dabb742SStefan Behrens } 27748dabb742SStefan Behrens 27758dabb742SStefan Behrens btrfs_close_extra_devices(fs_info, fs_devices, 1); 27768dabb742SStefan Behrens 2777c59021f8Sliubo ret = btrfs_init_space_info(fs_info); 2778c59021f8Sliubo if (ret) { 2779c59021f8Sliubo printk(KERN_ERR "Failed to initial space info: %d\n", ret); 2780c59021f8Sliubo goto fail_block_groups; 2781c59021f8Sliubo } 2782c59021f8Sliubo 27831b1d1f66SJosef Bacik ret = btrfs_read_block_groups(extent_root); 27841b1d1f66SJosef Bacik if (ret) { 27851b1d1f66SJosef Bacik printk(KERN_ERR "Failed to read block groups: %d\n", ret); 27861b1d1f66SJosef Bacik goto fail_block_groups; 27871b1d1f66SJosef Bacik } 27885af3e8ccSStefan Behrens fs_info->num_tolerated_disk_barrier_failures = 27895af3e8ccSStefan Behrens btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); 2790292fd7fcSStefan Behrens if (fs_info->fs_devices->missing_devices > 2791292fd7fcSStefan Behrens fs_info->num_tolerated_disk_barrier_failures && 2792292fd7fcSStefan Behrens !(sb->s_flags & MS_RDONLY)) { 2793292fd7fcSStefan Behrens printk(KERN_WARNING 2794292fd7fcSStefan Behrens "Btrfs: too many missing devices, writeable mount is not allowed\n"); 2795292fd7fcSStefan Behrens goto fail_block_groups; 2796292fd7fcSStefan Behrens } 27979078a3e1SChris Mason 2798a74a4b97SChris Mason fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 2799a74a4b97SChris Mason "btrfs-cleaner"); 280057506d50SQinghuang Feng if (IS_ERR(fs_info->cleaner_kthread)) 28011b1d1f66SJosef Bacik goto fail_block_groups; 2802a74a4b97SChris Mason 2803a74a4b97SChris Mason fs_info->transaction_kthread = kthread_run(transaction_kthread, 2804a74a4b97SChris Mason tree_root, 2805a74a4b97SChris Mason "btrfs-transaction"); 280657506d50SQinghuang Feng if (IS_ERR(fs_info->transaction_kthread)) 28073f157a2fSChris Mason goto fail_cleaner; 2808a74a4b97SChris Mason 2809c289811cSChris Mason if (!btrfs_test_opt(tree_root, SSD) && 2810c289811cSChris Mason !btrfs_test_opt(tree_root, NOSSD) && 2811c289811cSChris Mason !fs_info->fs_devices->rotating) { 2812c289811cSChris Mason printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD " 2813c289811cSChris Mason "mode\n"); 2814c289811cSChris Mason btrfs_set_opt(fs_info->mount_opt, SSD); 2815c289811cSChris Mason } 2816c289811cSChris Mason 281721adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 281821adbd5cSStefan Behrens if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) { 281921adbd5cSStefan Behrens ret = btrfsic_mount(tree_root, fs_devices, 282021adbd5cSStefan Behrens btrfs_test_opt(tree_root, 282121adbd5cSStefan Behrens CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 282221adbd5cSStefan Behrens 1 : 0, 282321adbd5cSStefan Behrens fs_info->check_integrity_print_mask); 282421adbd5cSStefan Behrens if (ret) 282521adbd5cSStefan Behrens printk(KERN_WARNING "btrfs: failed to initialize" 282621adbd5cSStefan Behrens " integrity check module %s\n", sb->s_id); 282721adbd5cSStefan Behrens } 282821adbd5cSStefan Behrens #endif 2829bcef60f2SArne Jansen ret = btrfs_read_qgroup_config(fs_info); 2830bcef60f2SArne Jansen if (ret) 2831bcef60f2SArne Jansen goto fail_trans_kthread; 283221adbd5cSStefan Behrens 2833acce952bSliubo /* do not make disk changes in broken FS */ 283468ce9682SStefan Behrens if (btrfs_super_log_root(disk_super) != 0) { 2835e02119d5SChris Mason u64 bytenr = btrfs_super_log_root(disk_super); 2836d18a2c44SChris Mason 28377c2ca468SChris Mason if (fs_devices->rw_devices == 0) { 2838d397712bSChris Mason printk(KERN_WARNING "Btrfs log replay required " 2839d397712bSChris Mason "on RO media\n"); 28407c2ca468SChris Mason err = -EIO; 2841bcef60f2SArne Jansen goto fail_qgroup; 28427c2ca468SChris Mason } 2843e02119d5SChris Mason blocksize = 2844e02119d5SChris Mason btrfs_level_size(tree_root, 2845e02119d5SChris Mason btrfs_super_log_root_level(disk_super)); 2846e02119d5SChris Mason 28476f07e42eSAl Viro log_tree_root = btrfs_alloc_root(fs_info); 2848676e4c86SDan Carpenter if (!log_tree_root) { 2849676e4c86SDan Carpenter err = -ENOMEM; 2850bcef60f2SArne Jansen goto fail_qgroup; 2851676e4c86SDan Carpenter } 2852e02119d5SChris Mason 2853e02119d5SChris Mason __setup_root(nodesize, leafsize, sectorsize, stripesize, 2854e02119d5SChris Mason log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 2855e02119d5SChris Mason 2856e02119d5SChris Mason log_tree_root->node = read_tree_block(tree_root, bytenr, 285784234f3aSYan Zheng blocksize, 285884234f3aSYan Zheng generation + 1); 2859416bc658SJosef Bacik if (!log_tree_root->node || 2860416bc658SJosef Bacik !extent_buffer_uptodate(log_tree_root->node)) { 2861416bc658SJosef Bacik printk(KERN_ERR "btrfs: failed to read log tree\n"); 2862416bc658SJosef Bacik free_extent_buffer(log_tree_root->node); 2863416bc658SJosef Bacik kfree(log_tree_root); 2864416bc658SJosef Bacik goto fail_trans_kthread; 2865416bc658SJosef Bacik } 286679787eaaSJeff Mahoney /* returns with log_tree_root freed on success */ 2867e02119d5SChris Mason ret = btrfs_recover_log_trees(log_tree_root); 286879787eaaSJeff Mahoney if (ret) { 286979787eaaSJeff Mahoney btrfs_error(tree_root->fs_info, ret, 287079787eaaSJeff Mahoney "Failed to recover log tree"); 287179787eaaSJeff Mahoney free_extent_buffer(log_tree_root->node); 287279787eaaSJeff Mahoney kfree(log_tree_root); 287379787eaaSJeff Mahoney goto fail_trans_kthread; 287479787eaaSJeff Mahoney } 2875e556ce2cSYan Zheng 2876e556ce2cSYan Zheng if (sb->s_flags & MS_RDONLY) { 2877e556ce2cSYan Zheng ret = btrfs_commit_super(tree_root); 287879787eaaSJeff Mahoney if (ret) 287979787eaaSJeff Mahoney goto fail_trans_kthread; 2880e556ce2cSYan Zheng } 2881e02119d5SChris Mason } 28821a40e23bSZheng Yan 288376dda93cSYan, Zheng ret = btrfs_find_orphan_roots(tree_root); 288479787eaaSJeff Mahoney if (ret) 288579787eaaSJeff Mahoney goto fail_trans_kthread; 288676dda93cSYan, Zheng 28877c2ca468SChris Mason if (!(sb->s_flags & MS_RDONLY)) { 2888d68fc57bSYan, Zheng ret = btrfs_cleanup_fs_roots(fs_info); 288944c44af2SIlya Dryomov if (ret) 289044c44af2SIlya Dryomov goto fail_trans_kthread; 2891d68fc57bSYan, Zheng 28925d4f98a2SYan Zheng ret = btrfs_recover_relocation(tree_root); 2893d7ce5843SMiao Xie if (ret < 0) { 2894d7ce5843SMiao Xie printk(KERN_WARNING 2895d7ce5843SMiao Xie "btrfs: failed to recover relocation\n"); 2896d7ce5843SMiao Xie err = -EINVAL; 2897bcef60f2SArne Jansen goto fail_qgroup; 2898d7ce5843SMiao Xie } 28997c2ca468SChris Mason } 29001a40e23bSZheng Yan 29013de4586cSChris Mason location.objectid = BTRFS_FS_TREE_OBJECTID; 29023de4586cSChris Mason location.type = BTRFS_ROOT_ITEM_KEY; 2903cb517eabSMiao Xie location.offset = 0; 29043de4586cSChris Mason 29053de4586cSChris Mason fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 29063140c9a3SDan Carpenter if (IS_ERR(fs_info->fs_root)) { 29073140c9a3SDan Carpenter err = PTR_ERR(fs_info->fs_root); 2908bcef60f2SArne Jansen goto fail_qgroup; 29093140c9a3SDan Carpenter } 2910c289811cSChris Mason 29112b6ba629SIlya Dryomov if (sb->s_flags & MS_RDONLY) 29122b6ba629SIlya Dryomov return 0; 29132b6ba629SIlya Dryomov 2914e3acc2a6SJosef Bacik down_read(&fs_info->cleanup_work_sem); 29152b6ba629SIlya Dryomov if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 29162b6ba629SIlya Dryomov (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 29172b6ba629SIlya Dryomov up_read(&fs_info->cleanup_work_sem); 29182b6ba629SIlya Dryomov close_ctree(tree_root); 29192b6ba629SIlya Dryomov return ret; 29202b6ba629SIlya Dryomov } 2921e3acc2a6SJosef Bacik up_read(&fs_info->cleanup_work_sem); 292259641015SIlya Dryomov 29232b6ba629SIlya Dryomov ret = btrfs_resume_balance_async(fs_info); 29242b6ba629SIlya Dryomov if (ret) { 29252b6ba629SIlya Dryomov printk(KERN_WARNING "btrfs: failed to resume balance\n"); 292666b4ffd1SJosef Bacik close_ctree(tree_root); 29272b6ba629SIlya Dryomov return ret; 2928e3acc2a6SJosef Bacik } 2929e3acc2a6SJosef Bacik 29308dabb742SStefan Behrens ret = btrfs_resume_dev_replace_async(fs_info); 29318dabb742SStefan Behrens if (ret) { 29328dabb742SStefan Behrens pr_warn("btrfs: failed to resume dev_replace\n"); 29338dabb742SStefan Behrens close_ctree(tree_root); 29348dabb742SStefan Behrens return ret; 29358dabb742SStefan Behrens } 29368dabb742SStefan Behrens 2937b382a324SJan Schmidt btrfs_qgroup_rescan_resume(fs_info); 2938b382a324SJan Schmidt 2939f7a81ea4SStefan Behrens if (create_uuid_tree) { 2940f7a81ea4SStefan Behrens pr_info("btrfs: creating UUID tree\n"); 2941f7a81ea4SStefan Behrens ret = btrfs_create_uuid_tree(fs_info); 2942f7a81ea4SStefan Behrens if (ret) { 2943f7a81ea4SStefan Behrens pr_warn("btrfs: failed to create the UUID tree %d\n", 2944f7a81ea4SStefan Behrens ret); 2945f7a81ea4SStefan Behrens close_ctree(tree_root); 2946f7a81ea4SStefan Behrens return ret; 2947f7a81ea4SStefan Behrens } 2948f420ee1eSStefan Behrens } else if (check_uuid_tree || 2949f420ee1eSStefan Behrens btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) { 295070f80175SStefan Behrens pr_info("btrfs: checking UUID tree\n"); 295170f80175SStefan Behrens ret = btrfs_check_uuid_tree(fs_info); 295270f80175SStefan Behrens if (ret) { 295370f80175SStefan Behrens pr_warn("btrfs: failed to check the UUID tree %d\n", 295470f80175SStefan Behrens ret); 295570f80175SStefan Behrens close_ctree(tree_root); 295670f80175SStefan Behrens return ret; 295770f80175SStefan Behrens } 295870f80175SStefan Behrens } else { 295970f80175SStefan Behrens fs_info->update_uuid_tree_gen = 1; 2960f7a81ea4SStefan Behrens } 2961f7a81ea4SStefan Behrens 2962ad2b2c80SAl Viro return 0; 296339279cc3SChris Mason 2964bcef60f2SArne Jansen fail_qgroup: 2965bcef60f2SArne Jansen btrfs_free_qgroup_config(fs_info); 29667c2ca468SChris Mason fail_trans_kthread: 29677c2ca468SChris Mason kthread_stop(fs_info->transaction_kthread); 296854067ae9SJosef Bacik btrfs_cleanup_transaction(fs_info->tree_root); 29697b5ff90eSJosef Bacik del_fs_roots(fs_info); 29703f157a2fSChris Mason fail_cleaner: 2971a74a4b97SChris Mason kthread_stop(fs_info->cleaner_kthread); 29727c2ca468SChris Mason 29737c2ca468SChris Mason /* 29747c2ca468SChris Mason * make sure we're done with the btree inode before we stop our 29757c2ca468SChris Mason * kthreads 29767c2ca468SChris Mason */ 29777c2ca468SChris Mason filemap_write_and_wait(fs_info->btree_inode->i_mapping); 29787c2ca468SChris Mason 29791b1d1f66SJosef Bacik fail_block_groups: 298054067ae9SJosef Bacik btrfs_put_block_group_cache(fs_info); 29811b1d1f66SJosef Bacik btrfs_free_block_groups(fs_info); 2982af31f5e5SChris Mason 2983af31f5e5SChris Mason fail_tree_roots: 2984af31f5e5SChris Mason free_root_pointers(fs_info, 1); 29852b8195bbSMiao Xie invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 2986af31f5e5SChris Mason 298739279cc3SChris Mason fail_sb_buffer: 29887abadb64SLiu Bo btrfs_stop_all_workers(fs_info); 298916cdcec7SMiao Xie fail_alloc: 29904543df7eSChris Mason fail_iput: 2991586e46e2SIlya Dryomov btrfs_mapping_tree_free(&fs_info->mapping_tree); 2992586e46e2SIlya Dryomov 29934543df7eSChris Mason iput(fs_info->btree_inode); 2994963d678bSMiao Xie fail_delalloc_bytes: 2995963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 2996e2d84521SMiao Xie fail_dirty_metadata_bytes: 2997e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 2998ad081f14SJens Axboe fail_bdi: 29997e662854SQinghuang Feng bdi_destroy(&fs_info->bdi); 300076dda93cSYan, Zheng fail_srcu: 300176dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 30027e662854SQinghuang Feng fail: 300353b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 3004586e46e2SIlya Dryomov btrfs_close_devices(fs_info->fs_devices); 3005ad2b2c80SAl Viro return err; 3006af31f5e5SChris Mason 3007af31f5e5SChris Mason recovery_tree_root: 3008af31f5e5SChris Mason if (!btrfs_test_opt(tree_root, RECOVERY)) 3009af31f5e5SChris Mason goto fail_tree_roots; 3010af31f5e5SChris Mason 3011af31f5e5SChris Mason free_root_pointers(fs_info, 0); 3012af31f5e5SChris Mason 3013af31f5e5SChris Mason /* don't use the log in recovery mode, it won't be valid */ 3014af31f5e5SChris Mason btrfs_set_super_log_root(disk_super, 0); 3015af31f5e5SChris Mason 3016af31f5e5SChris Mason /* we can't trust the free space cache either */ 3017af31f5e5SChris Mason btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3018af31f5e5SChris Mason 3019af31f5e5SChris Mason ret = next_root_backup(fs_info, fs_info->super_copy, 3020af31f5e5SChris Mason &num_backups_tried, &backup_index); 3021af31f5e5SChris Mason if (ret == -1) 3022af31f5e5SChris Mason goto fail_block_groups; 3023af31f5e5SChris Mason goto retry_root_backup; 3024eb60ceacSChris Mason } 3025eb60ceacSChris Mason 3026f2984462SChris Mason static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3027f2984462SChris Mason { 3028f2984462SChris Mason if (uptodate) { 3029f2984462SChris Mason set_buffer_uptodate(bh); 3030f2984462SChris Mason } else { 3031442a4f63SStefan Behrens struct btrfs_device *device = (struct btrfs_device *) 3032442a4f63SStefan Behrens bh->b_private; 3033442a4f63SStefan Behrens 3034606686eeSJosef Bacik printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to " 3035606686eeSJosef Bacik "I/O error on %s\n", 3036606686eeSJosef Bacik rcu_str_deref(device->name)); 30371259ab75SChris Mason /* note, we dont' set_buffer_write_io_error because we have 30381259ab75SChris Mason * our own ways of dealing with the IO errors 30391259ab75SChris Mason */ 3040f2984462SChris Mason clear_buffer_uptodate(bh); 3041442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3042f2984462SChris Mason } 3043f2984462SChris Mason unlock_buffer(bh); 3044f2984462SChris Mason put_bh(bh); 3045f2984462SChris Mason } 3046f2984462SChris Mason 3047a512bbf8SYan Zheng struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3048a512bbf8SYan Zheng { 3049a512bbf8SYan Zheng struct buffer_head *bh; 3050a512bbf8SYan Zheng struct buffer_head *latest = NULL; 3051a512bbf8SYan Zheng struct btrfs_super_block *super; 3052a512bbf8SYan Zheng int i; 3053a512bbf8SYan Zheng u64 transid = 0; 3054a512bbf8SYan Zheng u64 bytenr; 3055a512bbf8SYan Zheng 3056a512bbf8SYan Zheng /* we would like to check all the supers, but that would make 3057a512bbf8SYan Zheng * a btrfs mount succeed after a mkfs from a different FS. 3058a512bbf8SYan Zheng * So, we need to add a special mount option to scan for 3059a512bbf8SYan Zheng * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3060a512bbf8SYan Zheng */ 3061a512bbf8SYan Zheng for (i = 0; i < 1; i++) { 3062a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 30638068a47eSAnand Jain if (bytenr + BTRFS_SUPER_INFO_SIZE >= 30648068a47eSAnand Jain i_size_read(bdev->bd_inode)) 3065a512bbf8SYan Zheng break; 30668068a47eSAnand Jain bh = __bread(bdev, bytenr / 4096, 30678068a47eSAnand Jain BTRFS_SUPER_INFO_SIZE); 3068a512bbf8SYan Zheng if (!bh) 3069a512bbf8SYan Zheng continue; 3070a512bbf8SYan Zheng 3071a512bbf8SYan Zheng super = (struct btrfs_super_block *)bh->b_data; 3072a512bbf8SYan Zheng if (btrfs_super_bytenr(super) != bytenr || 30733cae210fSQu Wenruo btrfs_super_magic(super) != BTRFS_MAGIC) { 3074a512bbf8SYan Zheng brelse(bh); 3075a512bbf8SYan Zheng continue; 3076a512bbf8SYan Zheng } 3077a512bbf8SYan Zheng 3078a512bbf8SYan Zheng if (!latest || btrfs_super_generation(super) > transid) { 3079a512bbf8SYan Zheng brelse(latest); 3080a512bbf8SYan Zheng latest = bh; 3081a512bbf8SYan Zheng transid = btrfs_super_generation(super); 3082a512bbf8SYan Zheng } else { 3083a512bbf8SYan Zheng brelse(bh); 3084a512bbf8SYan Zheng } 3085a512bbf8SYan Zheng } 3086a512bbf8SYan Zheng return latest; 3087a512bbf8SYan Zheng } 3088a512bbf8SYan Zheng 30894eedeb75SHisashi Hifumi /* 30904eedeb75SHisashi Hifumi * this should be called twice, once with wait == 0 and 30914eedeb75SHisashi Hifumi * once with wait == 1. When wait == 0 is done, all the buffer heads 30924eedeb75SHisashi Hifumi * we write are pinned. 30934eedeb75SHisashi Hifumi * 30944eedeb75SHisashi Hifumi * They are released when wait == 1 is done. 30954eedeb75SHisashi Hifumi * max_mirrors must be the same for both runs, and it indicates how 30964eedeb75SHisashi Hifumi * many supers on this one device should be written. 30974eedeb75SHisashi Hifumi * 30984eedeb75SHisashi Hifumi * max_mirrors == 0 means to write them all. 30994eedeb75SHisashi Hifumi */ 3100a512bbf8SYan Zheng static int write_dev_supers(struct btrfs_device *device, 3101a512bbf8SYan Zheng struct btrfs_super_block *sb, 3102a512bbf8SYan Zheng int do_barriers, int wait, int max_mirrors) 3103a512bbf8SYan Zheng { 3104a512bbf8SYan Zheng struct buffer_head *bh; 3105a512bbf8SYan Zheng int i; 3106a512bbf8SYan Zheng int ret; 3107a512bbf8SYan Zheng int errors = 0; 3108a512bbf8SYan Zheng u32 crc; 3109a512bbf8SYan Zheng u64 bytenr; 3110a512bbf8SYan Zheng 3111a512bbf8SYan Zheng if (max_mirrors == 0) 3112a512bbf8SYan Zheng max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3113a512bbf8SYan Zheng 3114a512bbf8SYan Zheng for (i = 0; i < max_mirrors; i++) { 3115a512bbf8SYan Zheng bytenr = btrfs_sb_offset(i); 3116a512bbf8SYan Zheng if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes) 3117a512bbf8SYan Zheng break; 3118a512bbf8SYan Zheng 3119a512bbf8SYan Zheng if (wait) { 3120a512bbf8SYan Zheng bh = __find_get_block(device->bdev, bytenr / 4096, 3121a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3122634554dcSJosef Bacik if (!bh) { 3123634554dcSJosef Bacik errors++; 3124634554dcSJosef Bacik continue; 3125634554dcSJosef Bacik } 3126a512bbf8SYan Zheng wait_on_buffer(bh); 31274eedeb75SHisashi Hifumi if (!buffer_uptodate(bh)) 31284eedeb75SHisashi Hifumi errors++; 31294eedeb75SHisashi Hifumi 31304eedeb75SHisashi Hifumi /* drop our reference */ 31314eedeb75SHisashi Hifumi brelse(bh); 31324eedeb75SHisashi Hifumi 31334eedeb75SHisashi Hifumi /* drop the reference from the wait == 0 run */ 3134a512bbf8SYan Zheng brelse(bh); 3135a512bbf8SYan Zheng continue; 3136a512bbf8SYan Zheng } else { 3137a512bbf8SYan Zheng btrfs_set_super_bytenr(sb, bytenr); 3138a512bbf8SYan Zheng 3139a512bbf8SYan Zheng crc = ~(u32)0; 3140b0496686SLiu Bo crc = btrfs_csum_data((char *)sb + 3141a512bbf8SYan Zheng BTRFS_CSUM_SIZE, crc, 3142a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE - 3143a512bbf8SYan Zheng BTRFS_CSUM_SIZE); 3144a512bbf8SYan Zheng btrfs_csum_final(crc, sb->csum); 3145a512bbf8SYan Zheng 31464eedeb75SHisashi Hifumi /* 31474eedeb75SHisashi Hifumi * one reference for us, and we leave it for the 31484eedeb75SHisashi Hifumi * caller 31494eedeb75SHisashi Hifumi */ 3150a512bbf8SYan Zheng bh = __getblk(device->bdev, bytenr / 4096, 3151a512bbf8SYan Zheng BTRFS_SUPER_INFO_SIZE); 3152634554dcSJosef Bacik if (!bh) { 3153634554dcSJosef Bacik printk(KERN_ERR "btrfs: couldn't get super " 3154634554dcSJosef Bacik "buffer head for bytenr %Lu\n", bytenr); 3155634554dcSJosef Bacik errors++; 3156634554dcSJosef Bacik continue; 3157634554dcSJosef Bacik } 3158634554dcSJosef Bacik 3159a512bbf8SYan Zheng memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3160a512bbf8SYan Zheng 31614eedeb75SHisashi Hifumi /* one reference for submit_bh */ 3162a512bbf8SYan Zheng get_bh(bh); 31634eedeb75SHisashi Hifumi 31644eedeb75SHisashi Hifumi set_buffer_uptodate(bh); 3165a512bbf8SYan Zheng lock_buffer(bh); 3166a512bbf8SYan Zheng bh->b_end_io = btrfs_end_buffer_write_sync; 3167442a4f63SStefan Behrens bh->b_private = device; 3168a512bbf8SYan Zheng } 3169a512bbf8SYan Zheng 3170387125fcSChris Mason /* 3171387125fcSChris Mason * we fua the first super. The others we allow 3172387125fcSChris Mason * to go down lazy. 3173387125fcSChris Mason */ 317421adbd5cSStefan Behrens ret = btrfsic_submit_bh(WRITE_FUA, bh); 31754eedeb75SHisashi Hifumi if (ret) 3176a512bbf8SYan Zheng errors++; 3177a512bbf8SYan Zheng } 3178a512bbf8SYan Zheng return errors < i ? 0 : -1; 3179a512bbf8SYan Zheng } 3180a512bbf8SYan Zheng 3181387125fcSChris Mason /* 3182387125fcSChris Mason * endio for the write_dev_flush, this will wake anyone waiting 3183387125fcSChris Mason * for the barrier when it is done 3184387125fcSChris Mason */ 3185387125fcSChris Mason static void btrfs_end_empty_barrier(struct bio *bio, int err) 3186387125fcSChris Mason { 3187387125fcSChris Mason if (err) { 3188387125fcSChris Mason if (err == -EOPNOTSUPP) 3189387125fcSChris Mason set_bit(BIO_EOPNOTSUPP, &bio->bi_flags); 3190387125fcSChris Mason clear_bit(BIO_UPTODATE, &bio->bi_flags); 3191387125fcSChris Mason } 3192387125fcSChris Mason if (bio->bi_private) 3193387125fcSChris Mason complete(bio->bi_private); 3194387125fcSChris Mason bio_put(bio); 3195387125fcSChris Mason } 3196387125fcSChris Mason 3197387125fcSChris Mason /* 3198387125fcSChris Mason * trigger flushes for one the devices. If you pass wait == 0, the flushes are 3199387125fcSChris Mason * sent down. With wait == 1, it waits for the previous flush. 3200387125fcSChris Mason * 3201387125fcSChris Mason * any device where the flush fails with eopnotsupp are flagged as not-barrier 3202387125fcSChris Mason * capable 3203387125fcSChris Mason */ 3204387125fcSChris Mason static int write_dev_flush(struct btrfs_device *device, int wait) 3205387125fcSChris Mason { 3206387125fcSChris Mason struct bio *bio; 3207387125fcSChris Mason int ret = 0; 3208387125fcSChris Mason 3209387125fcSChris Mason if (device->nobarriers) 3210387125fcSChris Mason return 0; 3211387125fcSChris Mason 3212387125fcSChris Mason if (wait) { 3213387125fcSChris Mason bio = device->flush_bio; 3214387125fcSChris Mason if (!bio) 3215387125fcSChris Mason return 0; 3216387125fcSChris Mason 3217387125fcSChris Mason wait_for_completion(&device->flush_wait); 3218387125fcSChris Mason 3219387125fcSChris Mason if (bio_flagged(bio, BIO_EOPNOTSUPP)) { 3220606686eeSJosef Bacik printk_in_rcu("btrfs: disabling barriers on dev %s\n", 3221606686eeSJosef Bacik rcu_str_deref(device->name)); 3222387125fcSChris Mason device->nobarriers = 1; 32235af3e8ccSStefan Behrens } else if (!bio_flagged(bio, BIO_UPTODATE)) { 3224387125fcSChris Mason ret = -EIO; 3225442a4f63SStefan Behrens btrfs_dev_stat_inc_and_print(device, 3226442a4f63SStefan Behrens BTRFS_DEV_STAT_FLUSH_ERRS); 3227387125fcSChris Mason } 3228387125fcSChris Mason 3229387125fcSChris Mason /* drop the reference from the wait == 0 run */ 3230387125fcSChris Mason bio_put(bio); 3231387125fcSChris Mason device->flush_bio = NULL; 3232387125fcSChris Mason 3233387125fcSChris Mason return ret; 3234387125fcSChris Mason } 3235387125fcSChris Mason 3236387125fcSChris Mason /* 3237387125fcSChris Mason * one reference for us, and we leave it for the 3238387125fcSChris Mason * caller 3239387125fcSChris Mason */ 32409c017abcSJesper Juhl device->flush_bio = NULL; 32419be3395bSChris Mason bio = btrfs_io_bio_alloc(GFP_NOFS, 0); 3242387125fcSChris Mason if (!bio) 3243387125fcSChris Mason return -ENOMEM; 3244387125fcSChris Mason 3245387125fcSChris Mason bio->bi_end_io = btrfs_end_empty_barrier; 3246387125fcSChris Mason bio->bi_bdev = device->bdev; 3247387125fcSChris Mason init_completion(&device->flush_wait); 3248387125fcSChris Mason bio->bi_private = &device->flush_wait; 3249387125fcSChris Mason device->flush_bio = bio; 3250387125fcSChris Mason 3251387125fcSChris Mason bio_get(bio); 325221adbd5cSStefan Behrens btrfsic_submit_bio(WRITE_FLUSH, bio); 3253387125fcSChris Mason 3254387125fcSChris Mason return 0; 3255387125fcSChris Mason } 3256387125fcSChris Mason 3257387125fcSChris Mason /* 3258387125fcSChris Mason * send an empty flush down to each device in parallel, 3259387125fcSChris Mason * then wait for them 3260387125fcSChris Mason */ 3261387125fcSChris Mason static int barrier_all_devices(struct btrfs_fs_info *info) 3262387125fcSChris Mason { 3263387125fcSChris Mason struct list_head *head; 3264387125fcSChris Mason struct btrfs_device *dev; 32655af3e8ccSStefan Behrens int errors_send = 0; 32665af3e8ccSStefan Behrens int errors_wait = 0; 3267387125fcSChris Mason int ret; 3268387125fcSChris Mason 3269387125fcSChris Mason /* send down all the barriers */ 3270387125fcSChris Mason head = &info->fs_devices->devices; 3271387125fcSChris Mason list_for_each_entry_rcu(dev, head, dev_list) { 3272387125fcSChris Mason if (!dev->bdev) { 32735af3e8ccSStefan Behrens errors_send++; 3274387125fcSChris Mason continue; 3275387125fcSChris Mason } 3276387125fcSChris Mason if (!dev->in_fs_metadata || !dev->writeable) 3277387125fcSChris Mason continue; 3278387125fcSChris Mason 3279387125fcSChris Mason ret = write_dev_flush(dev, 0); 3280387125fcSChris Mason if (ret) 32815af3e8ccSStefan Behrens errors_send++; 3282387125fcSChris Mason } 3283387125fcSChris Mason 3284387125fcSChris Mason /* wait for all the barriers */ 3285387125fcSChris Mason list_for_each_entry_rcu(dev, head, dev_list) { 3286387125fcSChris Mason if (!dev->bdev) { 32875af3e8ccSStefan Behrens errors_wait++; 3288387125fcSChris Mason continue; 3289387125fcSChris Mason } 3290387125fcSChris Mason if (!dev->in_fs_metadata || !dev->writeable) 3291387125fcSChris Mason continue; 3292387125fcSChris Mason 3293387125fcSChris Mason ret = write_dev_flush(dev, 1); 3294387125fcSChris Mason if (ret) 32955af3e8ccSStefan Behrens errors_wait++; 3296387125fcSChris Mason } 32975af3e8ccSStefan Behrens if (errors_send > info->num_tolerated_disk_barrier_failures || 32985af3e8ccSStefan Behrens errors_wait > info->num_tolerated_disk_barrier_failures) 3299387125fcSChris Mason return -EIO; 3300387125fcSChris Mason return 0; 3301387125fcSChris Mason } 3302387125fcSChris Mason 33035af3e8ccSStefan Behrens int btrfs_calc_num_tolerated_disk_barrier_failures( 33045af3e8ccSStefan Behrens struct btrfs_fs_info *fs_info) 33055af3e8ccSStefan Behrens { 33065af3e8ccSStefan Behrens struct btrfs_ioctl_space_info space; 33075af3e8ccSStefan Behrens struct btrfs_space_info *sinfo; 33085af3e8ccSStefan Behrens u64 types[] = {BTRFS_BLOCK_GROUP_DATA, 33095af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_SYSTEM, 33105af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_METADATA, 33115af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA}; 33125af3e8ccSStefan Behrens int num_types = 4; 33135af3e8ccSStefan Behrens int i; 33145af3e8ccSStefan Behrens int c; 33155af3e8ccSStefan Behrens int num_tolerated_disk_barrier_failures = 33165af3e8ccSStefan Behrens (int)fs_info->fs_devices->num_devices; 33175af3e8ccSStefan Behrens 33185af3e8ccSStefan Behrens for (i = 0; i < num_types; i++) { 33195af3e8ccSStefan Behrens struct btrfs_space_info *tmp; 33205af3e8ccSStefan Behrens 33215af3e8ccSStefan Behrens sinfo = NULL; 33225af3e8ccSStefan Behrens rcu_read_lock(); 33235af3e8ccSStefan Behrens list_for_each_entry_rcu(tmp, &fs_info->space_info, list) { 33245af3e8ccSStefan Behrens if (tmp->flags == types[i]) { 33255af3e8ccSStefan Behrens sinfo = tmp; 33265af3e8ccSStefan Behrens break; 33275af3e8ccSStefan Behrens } 33285af3e8ccSStefan Behrens } 33295af3e8ccSStefan Behrens rcu_read_unlock(); 33305af3e8ccSStefan Behrens 33315af3e8ccSStefan Behrens if (!sinfo) 33325af3e8ccSStefan Behrens continue; 33335af3e8ccSStefan Behrens 33345af3e8ccSStefan Behrens down_read(&sinfo->groups_sem); 33355af3e8ccSStefan Behrens for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 33365af3e8ccSStefan Behrens if (!list_empty(&sinfo->block_groups[c])) { 33375af3e8ccSStefan Behrens u64 flags; 33385af3e8ccSStefan Behrens 33395af3e8ccSStefan Behrens btrfs_get_block_group_info( 33405af3e8ccSStefan Behrens &sinfo->block_groups[c], &space); 33415af3e8ccSStefan Behrens if (space.total_bytes == 0 || 33425af3e8ccSStefan Behrens space.used_bytes == 0) 33435af3e8ccSStefan Behrens continue; 33445af3e8ccSStefan Behrens flags = space.flags; 33455af3e8ccSStefan Behrens /* 33465af3e8ccSStefan Behrens * return 33475af3e8ccSStefan Behrens * 0: if dup, single or RAID0 is configured for 33485af3e8ccSStefan Behrens * any of metadata, system or data, else 33495af3e8ccSStefan Behrens * 1: if RAID5 is configured, or if RAID1 or 33505af3e8ccSStefan Behrens * RAID10 is configured and only two mirrors 33515af3e8ccSStefan Behrens * are used, else 33525af3e8ccSStefan Behrens * 2: if RAID6 is configured, else 33535af3e8ccSStefan Behrens * num_mirrors - 1: if RAID1 or RAID10 is 33545af3e8ccSStefan Behrens * configured and more than 33555af3e8ccSStefan Behrens * 2 mirrors are used. 33565af3e8ccSStefan Behrens */ 33575af3e8ccSStefan Behrens if (num_tolerated_disk_barrier_failures > 0 && 33585af3e8ccSStefan Behrens ((flags & (BTRFS_BLOCK_GROUP_DUP | 33595af3e8ccSStefan Behrens BTRFS_BLOCK_GROUP_RAID0)) || 33605af3e8ccSStefan Behrens ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) 33615af3e8ccSStefan Behrens == 0))) 33625af3e8ccSStefan Behrens num_tolerated_disk_barrier_failures = 0; 336353b381b3SDavid Woodhouse else if (num_tolerated_disk_barrier_failures > 1) { 336453b381b3SDavid Woodhouse if (flags & (BTRFS_BLOCK_GROUP_RAID1 | 336553b381b3SDavid Woodhouse BTRFS_BLOCK_GROUP_RAID5 | 336653b381b3SDavid Woodhouse BTRFS_BLOCK_GROUP_RAID10)) { 33675af3e8ccSStefan Behrens num_tolerated_disk_barrier_failures = 1; 336853b381b3SDavid Woodhouse } else if (flags & 336915b0a89dSHenrik Nordvik BTRFS_BLOCK_GROUP_RAID6) { 337053b381b3SDavid Woodhouse num_tolerated_disk_barrier_failures = 2; 337153b381b3SDavid Woodhouse } 337253b381b3SDavid Woodhouse } 33735af3e8ccSStefan Behrens } 33745af3e8ccSStefan Behrens } 33755af3e8ccSStefan Behrens up_read(&sinfo->groups_sem); 33765af3e8ccSStefan Behrens } 33775af3e8ccSStefan Behrens 33785af3e8ccSStefan Behrens return num_tolerated_disk_barrier_failures; 33795af3e8ccSStefan Behrens } 33805af3e8ccSStefan Behrens 338148a3b636SEric Sandeen static int write_all_supers(struct btrfs_root *root, int max_mirrors) 3382f2984462SChris Mason { 3383e5e9a520SChris Mason struct list_head *head; 3384f2984462SChris Mason struct btrfs_device *dev; 3385a061fc8dSChris Mason struct btrfs_super_block *sb; 3386f2984462SChris Mason struct btrfs_dev_item *dev_item; 3387f2984462SChris Mason int ret; 3388f2984462SChris Mason int do_barriers; 3389a236aed1SChris Mason int max_errors; 3390a236aed1SChris Mason int total_errors = 0; 3391a061fc8dSChris Mason u64 flags; 3392f2984462SChris Mason 3393f2984462SChris Mason do_barriers = !btrfs_test_opt(root, NOBARRIER); 3394af31f5e5SChris Mason backup_super_roots(root->fs_info); 3395f2984462SChris Mason 33966c41761fSDavid Sterba sb = root->fs_info->super_for_commit; 3397a061fc8dSChris Mason dev_item = &sb->dev_item; 3398e5e9a520SChris Mason 3399174ba509SChris Mason mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 3400e5e9a520SChris Mason head = &root->fs_info->fs_devices->devices; 3401d7306801SFilipe David Borba Manana max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1; 3402387125fcSChris Mason 34035af3e8ccSStefan Behrens if (do_barriers) { 34045af3e8ccSStefan Behrens ret = barrier_all_devices(root->fs_info); 34055af3e8ccSStefan Behrens if (ret) { 34065af3e8ccSStefan Behrens mutex_unlock( 34075af3e8ccSStefan Behrens &root->fs_info->fs_devices->device_list_mutex); 34085af3e8ccSStefan Behrens btrfs_error(root->fs_info, ret, 34095af3e8ccSStefan Behrens "errors while submitting device barriers."); 34105af3e8ccSStefan Behrens return ret; 34115af3e8ccSStefan Behrens } 34125af3e8ccSStefan Behrens } 3413387125fcSChris Mason 34141f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 3415dfe25020SChris Mason if (!dev->bdev) { 3416dfe25020SChris Mason total_errors++; 3417dfe25020SChris Mason continue; 3418dfe25020SChris Mason } 34192b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 3420dfe25020SChris Mason continue; 3421dfe25020SChris Mason 34222b82032cSYan Zheng btrfs_set_stack_device_generation(dev_item, 0); 3423a061fc8dSChris Mason btrfs_set_stack_device_type(dev_item, dev->type); 3424a061fc8dSChris Mason btrfs_set_stack_device_id(dev_item, dev->devid); 3425a061fc8dSChris Mason btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes); 3426a061fc8dSChris Mason btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used); 3427a061fc8dSChris Mason btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3428a061fc8dSChris Mason btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3429a061fc8dSChris Mason btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3430a061fc8dSChris Mason memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 34312b82032cSYan Zheng memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE); 3432a512bbf8SYan Zheng 3433a061fc8dSChris Mason flags = btrfs_super_flags(sb); 3434a061fc8dSChris Mason btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3435f2984462SChris Mason 3436a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors); 3437a236aed1SChris Mason if (ret) 3438a236aed1SChris Mason total_errors++; 3439f2984462SChris Mason } 3440a236aed1SChris Mason if (total_errors > max_errors) { 3441d397712bSChris Mason printk(KERN_ERR "btrfs: %d errors while writing supers\n", 3442d397712bSChris Mason total_errors); 3443a724b436SStefan Behrens mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 344479787eaaSJeff Mahoney 34459d565ba4SStefan Behrens /* FUA is masked off if unsupported and can't be the reason */ 34469d565ba4SStefan Behrens btrfs_error(root->fs_info, -EIO, 34479d565ba4SStefan Behrens "%d errors while writing supers", total_errors); 34489d565ba4SStefan Behrens return -EIO; 3449a236aed1SChris Mason } 3450f2984462SChris Mason 3451a512bbf8SYan Zheng total_errors = 0; 34521f78160cSXiao Guangrong list_for_each_entry_rcu(dev, head, dev_list) { 3453dfe25020SChris Mason if (!dev->bdev) 3454dfe25020SChris Mason continue; 34552b82032cSYan Zheng if (!dev->in_fs_metadata || !dev->writeable) 3456dfe25020SChris Mason continue; 3457dfe25020SChris Mason 3458a512bbf8SYan Zheng ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors); 3459a512bbf8SYan Zheng if (ret) 34601259ab75SChris Mason total_errors++; 3461f2984462SChris Mason } 3462174ba509SChris Mason mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 3463a236aed1SChris Mason if (total_errors > max_errors) { 346479787eaaSJeff Mahoney btrfs_error(root->fs_info, -EIO, 346579787eaaSJeff Mahoney "%d errors while writing supers", total_errors); 346679787eaaSJeff Mahoney return -EIO; 3467a236aed1SChris Mason } 3468f2984462SChris Mason return 0; 3469f2984462SChris Mason } 3470f2984462SChris Mason 3471a512bbf8SYan Zheng int write_ctree_super(struct btrfs_trans_handle *trans, 3472a512bbf8SYan Zheng struct btrfs_root *root, int max_mirrors) 3473cfaa7295SChris Mason { 3474e66f709bSChris Mason int ret; 34752c90e5d6SChris Mason 3476a512bbf8SYan Zheng ret = write_all_supers(root, max_mirrors); 34775f39d397SChris Mason return ret; 3478cfaa7295SChris Mason } 3479cfaa7295SChris Mason 3480cb517eabSMiao Xie /* Drop a fs root from the radix tree and free it. */ 3481cb517eabSMiao Xie void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3482cb517eabSMiao Xie struct btrfs_root *root) 34832619ba1fSChris Mason { 34844df27c4dSYan, Zheng spin_lock(&fs_info->fs_roots_radix_lock); 34852619ba1fSChris Mason radix_tree_delete(&fs_info->fs_roots_radix, 34862619ba1fSChris Mason (unsigned long)root->root_key.objectid); 34874df27c4dSYan, Zheng spin_unlock(&fs_info->fs_roots_radix_lock); 348876dda93cSYan, Zheng 348976dda93cSYan, Zheng if (btrfs_root_refs(&root->root_item) == 0) 349076dda93cSYan, Zheng synchronize_srcu(&fs_info->subvol_srcu); 349176dda93cSYan, Zheng 3492d7634482SLiu Bo if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 34933321719eSLiu Bo btrfs_free_log(NULL, root); 34943321719eSLiu Bo btrfs_free_log_root_tree(NULL, fs_info); 34953321719eSLiu Bo } 34963321719eSLiu Bo 3497581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_pinned); 3498581bb050SLi Zefan __btrfs_remove_free_space_cache(root->free_ino_ctl); 34994df27c4dSYan, Zheng free_fs_root(root); 35004df27c4dSYan, Zheng } 35014df27c4dSYan, Zheng 35024df27c4dSYan, Zheng static void free_fs_root(struct btrfs_root *root) 35034df27c4dSYan, Zheng { 350482d5902dSLi Zefan iput(root->cache_inode); 35054df27c4dSYan, Zheng WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 35061cb048f5SFilipe David Borba Manana btrfs_free_block_rsv(root, root->orphan_block_rsv); 35071cb048f5SFilipe David Borba Manana root->orphan_block_rsv = NULL; 35080ee5dc67SAl Viro if (root->anon_dev) 35090ee5dc67SAl Viro free_anon_bdev(root->anon_dev); 35105f39d397SChris Mason free_extent_buffer(root->node); 35115f39d397SChris Mason free_extent_buffer(root->commit_root); 3512581bb050SLi Zefan kfree(root->free_ino_ctl); 3513581bb050SLi Zefan kfree(root->free_ino_pinned); 351458176a96SJosef Bacik kfree(root->name); 3515b0feb9d9SMiao Xie btrfs_put_fs_root(root); 35162619ba1fSChris Mason } 35172619ba1fSChris Mason 3518cb517eabSMiao Xie void btrfs_free_fs_root(struct btrfs_root *root) 3519cb517eabSMiao Xie { 3520cb517eabSMiao Xie free_fs_root(root); 3521eb60ceacSChris Mason } 3522eb60ceacSChris Mason 3523c146afadSYan Zheng int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3524c146afadSYan Zheng { 3525c146afadSYan Zheng u64 root_objectid = 0; 3526c146afadSYan Zheng struct btrfs_root *gang[8]; 3527c146afadSYan Zheng int i; 3528c146afadSYan Zheng int ret; 3529c146afadSYan Zheng 3530c146afadSYan Zheng while (1) { 3531c146afadSYan Zheng ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3532c146afadSYan Zheng (void **)gang, root_objectid, 3533c146afadSYan Zheng ARRAY_SIZE(gang)); 3534c146afadSYan Zheng if (!ret) 3535c146afadSYan Zheng break; 35365d4f98a2SYan Zheng 35375d4f98a2SYan Zheng root_objectid = gang[ret - 1]->root_key.objectid + 1; 3538c146afadSYan Zheng for (i = 0; i < ret; i++) { 353966b4ffd1SJosef Bacik int err; 354066b4ffd1SJosef Bacik 3541c146afadSYan Zheng root_objectid = gang[i]->root_key.objectid; 354266b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(gang[i]); 354366b4ffd1SJosef Bacik if (err) 354466b4ffd1SJosef Bacik return err; 3545c146afadSYan Zheng } 3546c146afadSYan Zheng root_objectid++; 3547c146afadSYan Zheng } 3548c146afadSYan Zheng return 0; 3549c146afadSYan Zheng } 3550c146afadSYan Zheng 3551c146afadSYan Zheng int btrfs_commit_super(struct btrfs_root *root) 3552c146afadSYan Zheng { 3553c146afadSYan Zheng struct btrfs_trans_handle *trans; 3554c146afadSYan Zheng int ret; 3555c146afadSYan Zheng 3556c146afadSYan Zheng mutex_lock(&root->fs_info->cleaner_mutex); 355724bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 3558c146afadSYan Zheng mutex_unlock(&root->fs_info->cleaner_mutex); 35599d1a2a3aSDavid Sterba wake_up_process(root->fs_info->cleaner_kthread); 3560c71bf099SYan, Zheng 3561c71bf099SYan, Zheng /* wait until ongoing cleanup work done */ 3562c71bf099SYan, Zheng down_write(&root->fs_info->cleanup_work_sem); 3563c71bf099SYan, Zheng up_write(&root->fs_info->cleanup_work_sem); 3564c71bf099SYan, Zheng 35657a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 35663612b495STsutomu Itoh if (IS_ERR(trans)) 35673612b495STsutomu Itoh return PTR_ERR(trans); 3568c146afadSYan Zheng ret = btrfs_commit_transaction(trans, root); 356979787eaaSJeff Mahoney if (ret) 357079787eaaSJeff Mahoney return ret; 3571c146afadSYan Zheng /* run commit again to drop the original snapshot */ 35727a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 35733612b495STsutomu Itoh if (IS_ERR(trans)) 35743612b495STsutomu Itoh return PTR_ERR(trans); 357579787eaaSJeff Mahoney ret = btrfs_commit_transaction(trans, root); 357679787eaaSJeff Mahoney if (ret) 357779787eaaSJeff Mahoney return ret; 3578c146afadSYan Zheng ret = btrfs_write_and_wait_transaction(NULL, root); 357979787eaaSJeff Mahoney if (ret) { 358079787eaaSJeff Mahoney btrfs_error(root->fs_info, ret, 358179787eaaSJeff Mahoney "Failed to sync btree inode to disk."); 358279787eaaSJeff Mahoney return ret; 358379787eaaSJeff Mahoney } 3584c146afadSYan Zheng 3585a512bbf8SYan Zheng ret = write_ctree_super(NULL, root, 0); 3586c146afadSYan Zheng return ret; 3587c146afadSYan Zheng } 3588c146afadSYan Zheng 3589e20d96d6SChris Mason int close_ctree(struct btrfs_root *root) 3590eb60ceacSChris Mason { 35910f7d52f4SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 3592c146afadSYan Zheng int ret; 3593e089f05cSChris Mason 3594facda1e7SChris Mason fs_info->closing = 1; 3595a2135011SChris Mason smp_mb(); 3596a2135011SChris Mason 3597803b2f54SStefan Behrens /* wait for the uuid_scan task to finish */ 3598803b2f54SStefan Behrens down(&fs_info->uuid_tree_rescan_sem); 3599803b2f54SStefan Behrens /* avoid complains from lockdep et al., set sem back to initial state */ 3600803b2f54SStefan Behrens up(&fs_info->uuid_tree_rescan_sem); 3601803b2f54SStefan Behrens 3602837d5b6eSIlya Dryomov /* pause restriper - we want to resume on mount */ 3603aa1b8cd4SStefan Behrens btrfs_pause_balance(fs_info); 3604837d5b6eSIlya Dryomov 36058dabb742SStefan Behrens btrfs_dev_replace_suspend_for_unmount(fs_info); 36068dabb742SStefan Behrens 3607aa1b8cd4SStefan Behrens btrfs_scrub_cancel(fs_info); 36084cb5300bSChris Mason 36094cb5300bSChris Mason /* wait for any defraggers to finish */ 36104cb5300bSChris Mason wait_event(fs_info->transaction_wait, 36114cb5300bSChris Mason (atomic_read(&fs_info->defrag_running) == 0)); 36124cb5300bSChris Mason 36134cb5300bSChris Mason /* clear out the rbtree of defraggable inodes */ 361426176e7cSMiao Xie btrfs_cleanup_defrag_inodes(fs_info); 36154cb5300bSChris Mason 3616c146afadSYan Zheng if (!(fs_info->sb->s_flags & MS_RDONLY)) { 3617c146afadSYan Zheng ret = btrfs_commit_super(root); 3618d397712bSChris Mason if (ret) 3619d397712bSChris Mason printk(KERN_ERR "btrfs: commit super ret %d\n", ret); 3620c146afadSYan Zheng } 3621ed2ff2cbSChris Mason 362287533c47SMiao Xie if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 362368ce9682SStefan Behrens btrfs_error_commit_super(root); 3624acce952bSliubo 3625300e4f8aSJosef Bacik btrfs_put_block_group_cache(fs_info); 3626300e4f8aSJosef Bacik 3627e3029d9fSAl Viro kthread_stop(fs_info->transaction_kthread); 3628e3029d9fSAl Viro kthread_stop(fs_info->cleaner_kthread); 36298929ecfaSYan, Zheng 3630f25784b3SYan Zheng fs_info->closing = 2; 3631f25784b3SYan Zheng smp_mb(); 3632f25784b3SYan Zheng 3633bcef60f2SArne Jansen btrfs_free_qgroup_config(root->fs_info); 3634bcef60f2SArne Jansen 3635963d678bSMiao Xie if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 3636963d678bSMiao Xie printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n", 3637963d678bSMiao Xie percpu_counter_sum(&fs_info->delalloc_bytes)); 3638b0c68f8bSChris Mason } 363931153d81SYan Zheng 36400f7d52f4SChris Mason del_fs_roots(fs_info); 3641d10c5f31SChris Mason 36422b1360daSJosef Bacik btrfs_free_block_groups(fs_info); 36432b1360daSJosef Bacik 364496192499SJosef Bacik btrfs_stop_all_workers(fs_info); 364596192499SJosef Bacik 364613e6c37bSJosef Bacik free_root_pointers(fs_info, 1); 36479ad6b7bcSChris Mason 364813e6c37bSJosef Bacik iput(fs_info->btree_inode); 3649d6bfde87SChris Mason 365021adbd5cSStefan Behrens #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 365121adbd5cSStefan Behrens if (btrfs_test_opt(root, CHECK_INTEGRITY)) 365221adbd5cSStefan Behrens btrfsic_unmount(root, fs_info->fs_devices); 365321adbd5cSStefan Behrens #endif 365421adbd5cSStefan Behrens 3655dfe25020SChris Mason btrfs_close_devices(fs_info->fs_devices); 36560b86a832SChris Mason btrfs_mapping_tree_free(&fs_info->mapping_tree); 3657b248a415SChris Mason 3658e2d84521SMiao Xie percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3659963d678bSMiao Xie percpu_counter_destroy(&fs_info->delalloc_bytes); 366004160088SChris Mason bdi_destroy(&fs_info->bdi); 366176dda93cSYan, Zheng cleanup_srcu_struct(&fs_info->subvol_srcu); 36620b86a832SChris Mason 366353b381b3SDavid Woodhouse btrfs_free_stripe_hash_table(fs_info); 366453b381b3SDavid Woodhouse 36651cb048f5SFilipe David Borba Manana btrfs_free_block_rsv(root, root->orphan_block_rsv); 36661cb048f5SFilipe David Borba Manana root->orphan_block_rsv = NULL; 36671cb048f5SFilipe David Borba Manana 3668eb60ceacSChris Mason return 0; 3669eb60ceacSChris Mason } 3670eb60ceacSChris Mason 3671b9fab919SChris Mason int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 3672b9fab919SChris Mason int atomic) 3673ccd467d6SChris Mason { 36741259ab75SChris Mason int ret; 3675727011e0SChris Mason struct inode *btree_inode = buf->pages[0]->mapping->host; 36761259ab75SChris Mason 36770b32f4bbSJosef Bacik ret = extent_buffer_uptodate(buf); 36781259ab75SChris Mason if (!ret) 36791259ab75SChris Mason return ret; 36801259ab75SChris Mason 36811259ab75SChris Mason ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 3682b9fab919SChris Mason parent_transid, atomic); 3683b9fab919SChris Mason if (ret == -EAGAIN) 3684b9fab919SChris Mason return ret; 36851259ab75SChris Mason return !ret; 36865f39d397SChris Mason } 36876702ed49SChris Mason 36885f39d397SChris Mason int btrfs_set_buffer_uptodate(struct extent_buffer *buf) 36895f39d397SChris Mason { 36900b32f4bbSJosef Bacik return set_extent_buffer_uptodate(buf); 36915f39d397SChris Mason } 36925f39d397SChris Mason 36935f39d397SChris Mason void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 36945f39d397SChris Mason { 369506ea65a3SJosef Bacik struct btrfs_root *root; 36965f39d397SChris Mason u64 transid = btrfs_header_generation(buf); 3697b9473439SChris Mason int was_dirty; 3698b4ce94deSChris Mason 369906ea65a3SJosef Bacik #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 370006ea65a3SJosef Bacik /* 370106ea65a3SJosef Bacik * This is a fast path so only do this check if we have sanity tests 370206ea65a3SJosef Bacik * enabled. Normal people shouldn't be marking dummy buffers as dirty 370306ea65a3SJosef Bacik * outside of the sanity tests. 370406ea65a3SJosef Bacik */ 370506ea65a3SJosef Bacik if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags))) 370606ea65a3SJosef Bacik return; 370706ea65a3SJosef Bacik #endif 370806ea65a3SJosef Bacik root = BTRFS_I(buf->pages[0]->mapping->host)->root; 3709b9447ef8SChris Mason btrfs_assert_tree_locked(buf); 371031b1a2bdSJulia Lawall if (transid != root->fs_info->generation) 371131b1a2bdSJulia Lawall WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, " 3712d397712bSChris Mason "found %llu running %llu\n", 3713c1c9ff7cSGeert Uytterhoeven buf->start, transid, root->fs_info->generation); 37140b32f4bbSJosef Bacik was_dirty = set_extent_buffer_dirty(buf); 3715e2d84521SMiao Xie if (!was_dirty) 3716e2d84521SMiao Xie __percpu_counter_add(&root->fs_info->dirty_metadata_bytes, 3717e2d84521SMiao Xie buf->len, 3718e2d84521SMiao Xie root->fs_info->dirty_metadata_batch); 3719eb60ceacSChris Mason } 3720eb60ceacSChris Mason 3721b53d3f5dSLiu Bo static void __btrfs_btree_balance_dirty(struct btrfs_root *root, 3722b53d3f5dSLiu Bo int flush_delayed) 372335b7e476SChris Mason { 3724188de649SChris Mason /* 3725188de649SChris Mason * looks as though older kernels can get into trouble with 3726188de649SChris Mason * this code, they end up stuck in balance_dirty_pages forever 3727188de649SChris Mason */ 3728e2d84521SMiao Xie int ret; 3729d6bfde87SChris Mason 37306933c02eSJens Axboe if (current->flags & PF_MEMALLOC) 3731d6bfde87SChris Mason return; 3732d6bfde87SChris Mason 3733b53d3f5dSLiu Bo if (flush_delayed) 373416cdcec7SMiao Xie btrfs_balance_delayed_items(root); 373516cdcec7SMiao Xie 3736e2d84521SMiao Xie ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes, 3737e2d84521SMiao Xie BTRFS_DIRTY_METADATA_THRESH); 3738e2d84521SMiao Xie if (ret > 0) { 3739d0e1d66bSNamjae Jeon balance_dirty_pages_ratelimited( 3740d0e1d66bSNamjae Jeon root->fs_info->btree_inode->i_mapping); 374116cdcec7SMiao Xie } 374216cdcec7SMiao Xie return; 374316cdcec7SMiao Xie } 374416cdcec7SMiao Xie 3745b53d3f5dSLiu Bo void btrfs_btree_balance_dirty(struct btrfs_root *root) 374616cdcec7SMiao Xie { 3747b53d3f5dSLiu Bo __btrfs_btree_balance_dirty(root, 1); 374835b7e476SChris Mason } 3749b53d3f5dSLiu Bo 3750b53d3f5dSLiu Bo void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root) 3751b53d3f5dSLiu Bo { 3752b53d3f5dSLiu Bo __btrfs_btree_balance_dirty(root, 0); 3753d6bfde87SChris Mason } 37546b80053dSChris Mason 3755ca7a79adSChris Mason int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 37566b80053dSChris Mason { 3757727011e0SChris Mason struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root; 37580b32f4bbSJosef Bacik return btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 37596b80053dSChris Mason } 37600da5468fSChris Mason 3761fcd1f065SDavid Sterba static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 3762acce952bSliubo int read_only) 3763acce952bSliubo { 37641104a885SDavid Sterba /* 37651104a885SDavid Sterba * Placeholder for checks 37661104a885SDavid Sterba */ 3767fcd1f065SDavid Sterba return 0; 3768fcd1f065SDavid Sterba } 3769fcd1f065SDavid Sterba 377048a3b636SEric Sandeen static void btrfs_error_commit_super(struct btrfs_root *root) 3771acce952bSliubo { 3772acce952bSliubo mutex_lock(&root->fs_info->cleaner_mutex); 3773acce952bSliubo btrfs_run_delayed_iputs(root); 3774acce952bSliubo mutex_unlock(&root->fs_info->cleaner_mutex); 3775acce952bSliubo 3776acce952bSliubo down_write(&root->fs_info->cleanup_work_sem); 3777acce952bSliubo up_write(&root->fs_info->cleanup_work_sem); 3778acce952bSliubo 3779acce952bSliubo /* cleanup FS via transaction */ 3780acce952bSliubo btrfs_cleanup_transaction(root); 3781acce952bSliubo } 3782acce952bSliubo 3783569e0f35SJosef Bacik static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t, 3784569e0f35SJosef Bacik struct btrfs_root *root) 3785acce952bSliubo { 3786acce952bSliubo struct btrfs_inode *btrfs_inode; 3787acce952bSliubo struct list_head splice; 3788acce952bSliubo 3789acce952bSliubo INIT_LIST_HEAD(&splice); 3790acce952bSliubo 3791acce952bSliubo mutex_lock(&root->fs_info->ordered_operations_mutex); 3792199c2a9cSMiao Xie spin_lock(&root->fs_info->ordered_root_lock); 3793acce952bSliubo 3794569e0f35SJosef Bacik list_splice_init(&t->ordered_operations, &splice); 3795acce952bSliubo while (!list_empty(&splice)) { 3796acce952bSliubo btrfs_inode = list_entry(splice.next, struct btrfs_inode, 3797acce952bSliubo ordered_operations); 3798acce952bSliubo 3799acce952bSliubo list_del_init(&btrfs_inode->ordered_operations); 3800199c2a9cSMiao Xie spin_unlock(&root->fs_info->ordered_root_lock); 3801acce952bSliubo 3802acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 3803b216cbfbSMiao Xie 3804199c2a9cSMiao Xie spin_lock(&root->fs_info->ordered_root_lock); 3805acce952bSliubo } 3806acce952bSliubo 3807199c2a9cSMiao Xie spin_unlock(&root->fs_info->ordered_root_lock); 3808acce952bSliubo mutex_unlock(&root->fs_info->ordered_operations_mutex); 3809acce952bSliubo } 3810acce952bSliubo 3811143bede5SJeff Mahoney static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 3812acce952bSliubo { 3813acce952bSliubo struct btrfs_ordered_extent *ordered; 3814acce952bSliubo 3815199c2a9cSMiao Xie spin_lock(&root->ordered_extent_lock); 3816779880efSJosef Bacik /* 3817779880efSJosef Bacik * This will just short circuit the ordered completion stuff which will 3818779880efSJosef Bacik * make sure the ordered extent gets properly cleaned up. 3819779880efSJosef Bacik */ 3820199c2a9cSMiao Xie list_for_each_entry(ordered, &root->ordered_extents, 3821779880efSJosef Bacik root_extent_list) 3822779880efSJosef Bacik set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 3823199c2a9cSMiao Xie spin_unlock(&root->ordered_extent_lock); 3824199c2a9cSMiao Xie } 3825199c2a9cSMiao Xie 3826199c2a9cSMiao Xie static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 3827199c2a9cSMiao Xie { 3828199c2a9cSMiao Xie struct btrfs_root *root; 3829199c2a9cSMiao Xie struct list_head splice; 3830199c2a9cSMiao Xie 3831199c2a9cSMiao Xie INIT_LIST_HEAD(&splice); 3832199c2a9cSMiao Xie 3833199c2a9cSMiao Xie spin_lock(&fs_info->ordered_root_lock); 3834199c2a9cSMiao Xie list_splice_init(&fs_info->ordered_roots, &splice); 3835199c2a9cSMiao Xie while (!list_empty(&splice)) { 3836199c2a9cSMiao Xie root = list_first_entry(&splice, struct btrfs_root, 3837199c2a9cSMiao Xie ordered_root); 38381de2cfdeSJosef Bacik list_move_tail(&root->ordered_root, 38391de2cfdeSJosef Bacik &fs_info->ordered_roots); 3840199c2a9cSMiao Xie 3841199c2a9cSMiao Xie btrfs_destroy_ordered_extents(root); 3842199c2a9cSMiao Xie 3843199c2a9cSMiao Xie cond_resched_lock(&fs_info->ordered_root_lock); 3844199c2a9cSMiao Xie } 3845199c2a9cSMiao Xie spin_unlock(&fs_info->ordered_root_lock); 3846acce952bSliubo } 3847acce952bSliubo 384835a3621bSStefan Behrens static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 3849acce952bSliubo struct btrfs_root *root) 3850acce952bSliubo { 3851acce952bSliubo struct rb_node *node; 3852acce952bSliubo struct btrfs_delayed_ref_root *delayed_refs; 3853acce952bSliubo struct btrfs_delayed_ref_node *ref; 3854acce952bSliubo int ret = 0; 3855acce952bSliubo 3856acce952bSliubo delayed_refs = &trans->delayed_refs; 3857acce952bSliubo 3858acce952bSliubo spin_lock(&delayed_refs->lock); 3859acce952bSliubo if (delayed_refs->num_entries == 0) { 3860cfece4dbSDavid Sterba spin_unlock(&delayed_refs->lock); 3861acce952bSliubo printk(KERN_INFO "delayed_refs has NO entry\n"); 3862acce952bSliubo return ret; 3863acce952bSliubo } 3864acce952bSliubo 3865b939d1abSJosef Bacik while ((node = rb_first(&delayed_refs->root)) != NULL) { 3866eb12db69SJosef Bacik struct btrfs_delayed_ref_head *head = NULL; 3867e78417d1SJosef Bacik bool pin_bytes = false; 3868acce952bSliubo 3869eb12db69SJosef Bacik ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); 3870acce952bSliubo atomic_set(&ref->refs, 1); 3871acce952bSliubo if (btrfs_delayed_ref_is_head(ref)) { 3872acce952bSliubo 3873acce952bSliubo head = btrfs_delayed_node_to_head(ref); 3874b939d1abSJosef Bacik if (!mutex_trylock(&head->mutex)) { 3875b939d1abSJosef Bacik atomic_inc(&ref->refs); 387679787eaaSJeff Mahoney spin_unlock(&delayed_refs->lock); 3877b939d1abSJosef Bacik 3878b939d1abSJosef Bacik /* Need to wait for the delayed ref to run */ 3879acce952bSliubo mutex_lock(&head->mutex); 3880b939d1abSJosef Bacik mutex_unlock(&head->mutex); 3881b939d1abSJosef Bacik btrfs_put_delayed_ref(ref); 3882b939d1abSJosef Bacik 3883e18fca73SJosef Bacik spin_lock(&delayed_refs->lock); 3884b939d1abSJosef Bacik continue; 3885b939d1abSJosef Bacik } 3886b939d1abSJosef Bacik 388754067ae9SJosef Bacik if (head->must_insert_reserved) 3888e78417d1SJosef Bacik pin_bytes = true; 388978a6184aSMiao Xie btrfs_free_delayed_extent_op(head->extent_op); 3890acce952bSliubo delayed_refs->num_heads--; 3891acce952bSliubo if (list_empty(&head->cluster)) 3892acce952bSliubo delayed_refs->num_heads_ready--; 3893acce952bSliubo list_del_init(&head->cluster); 3894acce952bSliubo } 3895eb12db69SJosef Bacik 3896b939d1abSJosef Bacik ref->in_tree = 0; 3897b939d1abSJosef Bacik rb_erase(&ref->rb_node, &delayed_refs->root); 3898b939d1abSJosef Bacik delayed_refs->num_entries--; 3899acce952bSliubo spin_unlock(&delayed_refs->lock); 3900e78417d1SJosef Bacik if (head) { 3901e78417d1SJosef Bacik if (pin_bytes) 3902e78417d1SJosef Bacik btrfs_pin_extent(root, ref->bytenr, 3903e78417d1SJosef Bacik ref->num_bytes, 1); 3904e78417d1SJosef Bacik mutex_unlock(&head->mutex); 3905e78417d1SJosef Bacik } 3906acce952bSliubo btrfs_put_delayed_ref(ref); 3907acce952bSliubo 3908acce952bSliubo cond_resched(); 3909acce952bSliubo spin_lock(&delayed_refs->lock); 3910acce952bSliubo } 3911acce952bSliubo 3912acce952bSliubo spin_unlock(&delayed_refs->lock); 3913acce952bSliubo 3914acce952bSliubo return ret; 3915acce952bSliubo } 3916acce952bSliubo 3917143bede5SJeff Mahoney static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 3918acce952bSliubo { 3919acce952bSliubo struct btrfs_inode *btrfs_inode; 3920acce952bSliubo struct list_head splice; 3921acce952bSliubo 3922acce952bSliubo INIT_LIST_HEAD(&splice); 3923acce952bSliubo 3924eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 3925eb73c1b7SMiao Xie list_splice_init(&root->delalloc_inodes, &splice); 3926acce952bSliubo 3927acce952bSliubo while (!list_empty(&splice)) { 3928eb73c1b7SMiao Xie btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 3929acce952bSliubo delalloc_inodes); 3930acce952bSliubo 3931acce952bSliubo list_del_init(&btrfs_inode->delalloc_inodes); 3932df0af1a5SMiao Xie clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, 3933df0af1a5SMiao Xie &btrfs_inode->runtime_flags); 3934eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 3935acce952bSliubo 3936acce952bSliubo btrfs_invalidate_inodes(btrfs_inode->root); 3937b216cbfbSMiao Xie 3938eb73c1b7SMiao Xie spin_lock(&root->delalloc_lock); 3939acce952bSliubo } 3940acce952bSliubo 3941eb73c1b7SMiao Xie spin_unlock(&root->delalloc_lock); 3942eb73c1b7SMiao Xie } 3943eb73c1b7SMiao Xie 3944eb73c1b7SMiao Xie static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 3945eb73c1b7SMiao Xie { 3946eb73c1b7SMiao Xie struct btrfs_root *root; 3947eb73c1b7SMiao Xie struct list_head splice; 3948eb73c1b7SMiao Xie 3949eb73c1b7SMiao Xie INIT_LIST_HEAD(&splice); 3950eb73c1b7SMiao Xie 3951eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 3952eb73c1b7SMiao Xie list_splice_init(&fs_info->delalloc_roots, &splice); 3953eb73c1b7SMiao Xie while (!list_empty(&splice)) { 3954eb73c1b7SMiao Xie root = list_first_entry(&splice, struct btrfs_root, 3955eb73c1b7SMiao Xie delalloc_root); 3956eb73c1b7SMiao Xie list_del_init(&root->delalloc_root); 3957eb73c1b7SMiao Xie root = btrfs_grab_fs_root(root); 3958eb73c1b7SMiao Xie BUG_ON(!root); 3959eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 3960eb73c1b7SMiao Xie 3961eb73c1b7SMiao Xie btrfs_destroy_delalloc_inodes(root); 3962eb73c1b7SMiao Xie btrfs_put_fs_root(root); 3963eb73c1b7SMiao Xie 3964eb73c1b7SMiao Xie spin_lock(&fs_info->delalloc_root_lock); 3965eb73c1b7SMiao Xie } 3966eb73c1b7SMiao Xie spin_unlock(&fs_info->delalloc_root_lock); 3967acce952bSliubo } 3968acce952bSliubo 3969acce952bSliubo static int btrfs_destroy_marked_extents(struct btrfs_root *root, 3970acce952bSliubo struct extent_io_tree *dirty_pages, 3971acce952bSliubo int mark) 3972acce952bSliubo { 3973acce952bSliubo int ret; 3974acce952bSliubo struct extent_buffer *eb; 3975acce952bSliubo u64 start = 0; 3976acce952bSliubo u64 end; 3977acce952bSliubo 3978acce952bSliubo while (1) { 3979acce952bSliubo ret = find_first_extent_bit(dirty_pages, start, &start, &end, 3980e6138876SJosef Bacik mark, NULL); 3981acce952bSliubo if (ret) 3982acce952bSliubo break; 3983acce952bSliubo 3984acce952bSliubo clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 3985acce952bSliubo while (start <= end) { 3986fd8b2b61SJosef Bacik eb = btrfs_find_tree_block(root, start, 3987fd8b2b61SJosef Bacik root->leafsize); 398869a85bd8SJosef Bacik start += root->leafsize; 3989fd8b2b61SJosef Bacik if (!eb) 3990acce952bSliubo continue; 3991fd8b2b61SJosef Bacik wait_on_extent_buffer_writeback(eb); 3992acce952bSliubo 3993fd8b2b61SJosef Bacik if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 3994fd8b2b61SJosef Bacik &eb->bflags)) 3995fd8b2b61SJosef Bacik clear_extent_buffer_dirty(eb); 3996fd8b2b61SJosef Bacik free_extent_buffer_stale(eb); 3997acce952bSliubo } 3998acce952bSliubo } 3999acce952bSliubo 4000acce952bSliubo return ret; 4001acce952bSliubo } 4002acce952bSliubo 4003acce952bSliubo static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 4004acce952bSliubo struct extent_io_tree *pinned_extents) 4005acce952bSliubo { 4006acce952bSliubo struct extent_io_tree *unpin; 4007acce952bSliubo u64 start; 4008acce952bSliubo u64 end; 4009acce952bSliubo int ret; 4010ed0eaa14SLiu Bo bool loop = true; 4011acce952bSliubo 4012acce952bSliubo unpin = pinned_extents; 4013ed0eaa14SLiu Bo again: 4014acce952bSliubo while (1) { 4015acce952bSliubo ret = find_first_extent_bit(unpin, 0, &start, &end, 4016e6138876SJosef Bacik EXTENT_DIRTY, NULL); 4017acce952bSliubo if (ret) 4018acce952bSliubo break; 4019acce952bSliubo 4020acce952bSliubo /* opt_discard */ 40215378e607SLi Dongyang if (btrfs_test_opt(root, DISCARD)) 40225378e607SLi Dongyang ret = btrfs_error_discard_extent(root, start, 40235378e607SLi Dongyang end + 1 - start, 40245378e607SLi Dongyang NULL); 4025acce952bSliubo 4026acce952bSliubo clear_extent_dirty(unpin, start, end, GFP_NOFS); 4027acce952bSliubo btrfs_error_unpin_extent_range(root, start, end); 4028acce952bSliubo cond_resched(); 4029acce952bSliubo } 4030acce952bSliubo 4031ed0eaa14SLiu Bo if (loop) { 4032ed0eaa14SLiu Bo if (unpin == &root->fs_info->freed_extents[0]) 4033ed0eaa14SLiu Bo unpin = &root->fs_info->freed_extents[1]; 4034ed0eaa14SLiu Bo else 4035ed0eaa14SLiu Bo unpin = &root->fs_info->freed_extents[0]; 4036ed0eaa14SLiu Bo loop = false; 4037ed0eaa14SLiu Bo goto again; 4038ed0eaa14SLiu Bo } 4039ed0eaa14SLiu Bo 4040acce952bSliubo return 0; 4041acce952bSliubo } 4042acce952bSliubo 404349b25e05SJeff Mahoney void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 404449b25e05SJeff Mahoney struct btrfs_root *root) 404549b25e05SJeff Mahoney { 4046724e2315SJosef Bacik btrfs_destroy_ordered_operations(cur_trans, root); 4047724e2315SJosef Bacik 404849b25e05SJeff Mahoney btrfs_destroy_delayed_refs(cur_trans, root); 404949b25e05SJeff Mahoney 40504a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 405149b25e05SJeff Mahoney wake_up(&root->fs_info->transaction_blocked_wait); 405249b25e05SJeff Mahoney 40534a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 405449b25e05SJeff Mahoney wake_up(&root->fs_info->transaction_wait); 405549b25e05SJeff Mahoney 405667cde344SMiao Xie btrfs_destroy_delayed_inodes(root); 405767cde344SMiao Xie btrfs_assert_delayed_root_empty(root); 405867cde344SMiao Xie 405949b25e05SJeff Mahoney btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages, 406049b25e05SJeff Mahoney EXTENT_DIRTY); 40616e841e32SLiu Bo btrfs_destroy_pinned_extent(root, 40626e841e32SLiu Bo root->fs_info->pinned_extents); 406349b25e05SJeff Mahoney 40644a9d8bdeSMiao Xie cur_trans->state =TRANS_STATE_COMPLETED; 40654a9d8bdeSMiao Xie wake_up(&cur_trans->commit_wait); 40664a9d8bdeSMiao Xie 406749b25e05SJeff Mahoney /* 406849b25e05SJeff Mahoney memset(cur_trans, 0, sizeof(*cur_trans)); 406949b25e05SJeff Mahoney kmem_cache_free(btrfs_transaction_cachep, cur_trans); 407049b25e05SJeff Mahoney */ 407149b25e05SJeff Mahoney } 407249b25e05SJeff Mahoney 407348a3b636SEric Sandeen static int btrfs_cleanup_transaction(struct btrfs_root *root) 4074acce952bSliubo { 4075acce952bSliubo struct btrfs_transaction *t; 4076acce952bSliubo 4077acce952bSliubo mutex_lock(&root->fs_info->transaction_kthread_mutex); 4078acce952bSliubo 4079a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4080724e2315SJosef Bacik while (!list_empty(&root->fs_info->trans_list)) { 4081724e2315SJosef Bacik t = list_first_entry(&root->fs_info->trans_list, 4082724e2315SJosef Bacik struct btrfs_transaction, list); 4083724e2315SJosef Bacik if (t->state >= TRANS_STATE_COMMIT_START) { 4084724e2315SJosef Bacik atomic_inc(&t->use_count); 4085724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4086724e2315SJosef Bacik btrfs_wait_for_commit(root, t->transid); 4087724e2315SJosef Bacik btrfs_put_transaction(t); 4088724e2315SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4089724e2315SJosef Bacik continue; 4090724e2315SJosef Bacik } 4091724e2315SJosef Bacik if (t == root->fs_info->running_transaction) { 4092724e2315SJosef Bacik t->state = TRANS_STATE_COMMIT_DOING; 4093724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4094724e2315SJosef Bacik /* 4095724e2315SJosef Bacik * We wait for 0 num_writers since we don't hold a trans 4096724e2315SJosef Bacik * handle open currently for this transaction. 4097724e2315SJosef Bacik */ 4098724e2315SJosef Bacik wait_event(t->writer_wait, 4099724e2315SJosef Bacik atomic_read(&t->num_writers) == 0); 4100724e2315SJosef Bacik } else { 4101724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4102724e2315SJosef Bacik } 4103724e2315SJosef Bacik btrfs_cleanup_one_transaction(t, root); 4104724e2315SJosef Bacik 4105724e2315SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4106724e2315SJosef Bacik if (t == root->fs_info->running_transaction) 4107ac673879SMiao Xie root->fs_info->running_transaction = NULL; 4108724e2315SJosef Bacik list_del_init(&t->list); 4109a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4110a4abeea4SJosef Bacik 4111724e2315SJosef Bacik btrfs_put_transaction(t); 4112724e2315SJosef Bacik trace_btrfs_transaction_commit(root); 4113724e2315SJosef Bacik spin_lock(&root->fs_info->trans_lock); 4114724e2315SJosef Bacik } 4115724e2315SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 4116199c2a9cSMiao Xie btrfs_destroy_all_ordered_extents(root->fs_info); 411767cde344SMiao Xie btrfs_destroy_delayed_inodes(root); 411867cde344SMiao Xie btrfs_assert_delayed_root_empty(root); 4119724e2315SJosef Bacik btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents); 4120eb73c1b7SMiao Xie btrfs_destroy_all_delalloc_inodes(root->fs_info); 4121acce952bSliubo mutex_unlock(&root->fs_info->transaction_kthread_mutex); 4122acce952bSliubo 4123acce952bSliubo return 0; 4124acce952bSliubo } 4125acce952bSliubo 4126d1310b2eSChris Mason static struct extent_io_ops btree_extent_io_ops = { 4127ce9adaa5SChris Mason .readpage_end_io_hook = btree_readpage_end_io_hook, 41284bb31e92SArne Jansen .readpage_io_failed_hook = btree_io_failed_hook, 41290b86a832SChris Mason .submit_bio_hook = btree_submit_bio_hook, 4130239b14b3SChris Mason /* note we're sharing with inode.c for the merge bio hook */ 4131239b14b3SChris Mason .merge_bio_hook = btrfs_merge_bio_hook, 41320da5468fSChris Mason }; 4133